xref: /openbmc/linux/mm/slub.c (revision 9a8f3203)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * SLUB: A slab allocator that limits cache line use instead of queuing
4  * objects in per cpu and per node lists.
5  *
6  * The allocator synchronizes using per slab locks or atomic operatios
7  * and only uses a centralized lock to manage a pool of partial slabs.
8  *
9  * (C) 2007 SGI, Christoph Lameter
10  * (C) 2011 Linux Foundation, Christoph Lameter
11  */
12 
13 #include <linux/mm.h>
14 #include <linux/swap.h> /* struct reclaim_state */
15 #include <linux/module.h>
16 #include <linux/bit_spinlock.h>
17 #include <linux/interrupt.h>
18 #include <linux/bitops.h>
19 #include <linux/slab.h>
20 #include "slab.h"
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/kasan.h>
24 #include <linux/cpu.h>
25 #include <linux/cpuset.h>
26 #include <linux/mempolicy.h>
27 #include <linux/ctype.h>
28 #include <linux/debugobjects.h>
29 #include <linux/kallsyms.h>
30 #include <linux/memory.h>
31 #include <linux/math64.h>
32 #include <linux/fault-inject.h>
33 #include <linux/stacktrace.h>
34 #include <linux/prefetch.h>
35 #include <linux/memcontrol.h>
36 #include <linux/random.h>
37 
38 #include <trace/events/kmem.h>
39 
40 #include "internal.h"
41 
42 /*
43  * Lock order:
44  *   1. slab_mutex (Global Mutex)
45  *   2. node->list_lock
46  *   3. slab_lock(page) (Only on some arches and for debugging)
47  *
48  *   slab_mutex
49  *
50  *   The role of the slab_mutex is to protect the list of all the slabs
51  *   and to synchronize major metadata changes to slab cache structures.
52  *
53  *   The slab_lock is only used for debugging and on arches that do not
54  *   have the ability to do a cmpxchg_double. It only protects:
55  *	A. page->freelist	-> List of object free in a page
56  *	B. page->inuse		-> Number of objects in use
57  *	C. page->objects	-> Number of objects in page
58  *	D. page->frozen		-> frozen state
59  *
60  *   If a slab is frozen then it is exempt from list management. It is not
61  *   on any list. The processor that froze the slab is the one who can
62  *   perform list operations on the page. Other processors may put objects
63  *   onto the freelist but the processor that froze the slab is the only
64  *   one that can retrieve the objects from the page's freelist.
65  *
66  *   The list_lock protects the partial and full list on each node and
67  *   the partial slab counter. If taken then no new slabs may be added or
68  *   removed from the lists nor make the number of partial slabs be modified.
69  *   (Note that the total number of slabs is an atomic value that may be
70  *   modified without taking the list lock).
71  *
72  *   The list_lock is a centralized lock and thus we avoid taking it as
73  *   much as possible. As long as SLUB does not have to handle partial
74  *   slabs, operations can continue without any centralized lock. F.e.
75  *   allocating a long series of objects that fill up slabs does not require
76  *   the list lock.
77  *   Interrupts are disabled during allocation and deallocation in order to
78  *   make the slab allocator safe to use in the context of an irq. In addition
79  *   interrupts are disabled to ensure that the processor does not change
80  *   while handling per_cpu slabs, due to kernel preemption.
81  *
82  * SLUB assigns one slab for allocation to each processor.
83  * Allocations only occur from these slabs called cpu slabs.
84  *
85  * Slabs with free elements are kept on a partial list and during regular
86  * operations no list for full slabs is used. If an object in a full slab is
87  * freed then the slab will show up again on the partial lists.
88  * We track full slabs for debugging purposes though because otherwise we
89  * cannot scan all objects.
90  *
91  * Slabs are freed when they become empty. Teardown and setup is
92  * minimal so we rely on the page allocators per cpu caches for
93  * fast frees and allocs.
94  *
95  * Overloading of page flags that are otherwise used for LRU management.
96  *
97  * PageActive 		The slab is frozen and exempt from list processing.
98  * 			This means that the slab is dedicated to a purpose
99  * 			such as satisfying allocations for a specific
100  * 			processor. Objects may be freed in the slab while
101  * 			it is frozen but slab_free will then skip the usual
102  * 			list operations. It is up to the processor holding
103  * 			the slab to integrate the slab into the slab lists
104  * 			when the slab is no longer needed.
105  *
106  * 			One use of this flag is to mark slabs that are
107  * 			used for allocations. Then such a slab becomes a cpu
108  * 			slab. The cpu slab may be equipped with an additional
109  * 			freelist that allows lockless access to
110  * 			free objects in addition to the regular freelist
111  * 			that requires the slab lock.
112  *
113  * PageError		Slab requires special handling due to debug
114  * 			options set. This moves	slab handling out of
115  * 			the fast path and disables lockless freelists.
116  */
117 
118 static inline int kmem_cache_debug(struct kmem_cache *s)
119 {
120 #ifdef CONFIG_SLUB_DEBUG
121 	return unlikely(s->flags & SLAB_DEBUG_FLAGS);
122 #else
123 	return 0;
124 #endif
125 }
126 
127 void *fixup_red_left(struct kmem_cache *s, void *p)
128 {
129 	if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
130 		p += s->red_left_pad;
131 
132 	return p;
133 }
134 
135 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
136 {
137 #ifdef CONFIG_SLUB_CPU_PARTIAL
138 	return !kmem_cache_debug(s);
139 #else
140 	return false;
141 #endif
142 }
143 
144 /*
145  * Issues still to be resolved:
146  *
147  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
148  *
149  * - Variable sizing of the per node arrays
150  */
151 
152 /* Enable to test recovery from slab corruption on boot */
153 #undef SLUB_RESILIENCY_TEST
154 
155 /* Enable to log cmpxchg failures */
156 #undef SLUB_DEBUG_CMPXCHG
157 
158 /*
159  * Mininum number of partial slabs. These will be left on the partial
160  * lists even if they are empty. kmem_cache_shrink may reclaim them.
161  */
162 #define MIN_PARTIAL 5
163 
164 /*
165  * Maximum number of desirable partial slabs.
166  * The existence of more partial slabs makes kmem_cache_shrink
167  * sort the partial list by the number of objects in use.
168  */
169 #define MAX_PARTIAL 10
170 
171 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
172 				SLAB_POISON | SLAB_STORE_USER)
173 
174 /*
175  * These debug flags cannot use CMPXCHG because there might be consistency
176  * issues when checking or reading debug information
177  */
178 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
179 				SLAB_TRACE)
180 
181 
182 /*
183  * Debugging flags that require metadata to be stored in the slab.  These get
184  * disabled when slub_debug=O is used and a cache's min order increases with
185  * metadata.
186  */
187 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
188 
189 #define OO_SHIFT	16
190 #define OO_MASK		((1 << OO_SHIFT) - 1)
191 #define MAX_OBJS_PER_PAGE	32767 /* since page.objects is u15 */
192 
193 /* Internal SLUB flags */
194 /* Poison object */
195 #define __OBJECT_POISON		((slab_flags_t __force)0x80000000U)
196 /* Use cmpxchg_double */
197 #define __CMPXCHG_DOUBLE	((slab_flags_t __force)0x40000000U)
198 
199 /*
200  * Tracking user of a slab.
201  */
202 #define TRACK_ADDRS_COUNT 16
203 struct track {
204 	unsigned long addr;	/* Called from address */
205 #ifdef CONFIG_STACKTRACE
206 	unsigned long addrs[TRACK_ADDRS_COUNT];	/* Called from address */
207 #endif
208 	int cpu;		/* Was running on cpu */
209 	int pid;		/* Pid context */
210 	unsigned long when;	/* When did the operation occur */
211 };
212 
213 enum track_item { TRACK_ALLOC, TRACK_FREE };
214 
215 #ifdef CONFIG_SYSFS
216 static int sysfs_slab_add(struct kmem_cache *);
217 static int sysfs_slab_alias(struct kmem_cache *, const char *);
218 static void memcg_propagate_slab_attrs(struct kmem_cache *s);
219 static void sysfs_slab_remove(struct kmem_cache *s);
220 #else
221 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
222 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
223 							{ return 0; }
224 static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
225 static inline void sysfs_slab_remove(struct kmem_cache *s) { }
226 #endif
227 
228 static inline void stat(const struct kmem_cache *s, enum stat_item si)
229 {
230 #ifdef CONFIG_SLUB_STATS
231 	/*
232 	 * The rmw is racy on a preemptible kernel but this is acceptable, so
233 	 * avoid this_cpu_add()'s irq-disable overhead.
234 	 */
235 	raw_cpu_inc(s->cpu_slab->stat[si]);
236 #endif
237 }
238 
239 /********************************************************************
240  * 			Core slab cache functions
241  *******************************************************************/
242 
243 /*
244  * Returns freelist pointer (ptr). With hardening, this is obfuscated
245  * with an XOR of the address where the pointer is held and a per-cache
246  * random number.
247  */
248 static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
249 				 unsigned long ptr_addr)
250 {
251 #ifdef CONFIG_SLAB_FREELIST_HARDENED
252 	/*
253 	 * When CONFIG_KASAN_SW_TAGS is enabled, ptr_addr might be tagged.
254 	 * Normally, this doesn't cause any issues, as both set_freepointer()
255 	 * and get_freepointer() are called with a pointer with the same tag.
256 	 * However, there are some issues with CONFIG_SLUB_DEBUG code. For
257 	 * example, when __free_slub() iterates over objects in a cache, it
258 	 * passes untagged pointers to check_object(). check_object() in turns
259 	 * calls get_freepointer() with an untagged pointer, which causes the
260 	 * freepointer to be restored incorrectly.
261 	 */
262 	return (void *)((unsigned long)ptr ^ s->random ^
263 			(unsigned long)kasan_reset_tag((void *)ptr_addr));
264 #else
265 	return ptr;
266 #endif
267 }
268 
269 /* Returns the freelist pointer recorded at location ptr_addr. */
270 static inline void *freelist_dereference(const struct kmem_cache *s,
271 					 void *ptr_addr)
272 {
273 	return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
274 			    (unsigned long)ptr_addr);
275 }
276 
277 static inline void *get_freepointer(struct kmem_cache *s, void *object)
278 {
279 	return freelist_dereference(s, object + s->offset);
280 }
281 
282 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
283 {
284 	prefetch(object + s->offset);
285 }
286 
287 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
288 {
289 	unsigned long freepointer_addr;
290 	void *p;
291 
292 	if (!debug_pagealloc_enabled())
293 		return get_freepointer(s, object);
294 
295 	freepointer_addr = (unsigned long)object + s->offset;
296 	probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p));
297 	return freelist_ptr(s, p, freepointer_addr);
298 }
299 
300 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
301 {
302 	unsigned long freeptr_addr = (unsigned long)object + s->offset;
303 
304 #ifdef CONFIG_SLAB_FREELIST_HARDENED
305 	BUG_ON(object == fp); /* naive detection of double free or corruption */
306 #endif
307 
308 	*(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
309 }
310 
311 /* Loop over all objects in a slab */
312 #define for_each_object(__p, __s, __addr, __objects) \
313 	for (__p = fixup_red_left(__s, __addr); \
314 		__p < (__addr) + (__objects) * (__s)->size; \
315 		__p += (__s)->size)
316 
317 /* Determine object index from a given position */
318 static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr)
319 {
320 	return (kasan_reset_tag(p) - addr) / s->size;
321 }
322 
323 static inline unsigned int order_objects(unsigned int order, unsigned int size)
324 {
325 	return ((unsigned int)PAGE_SIZE << order) / size;
326 }
327 
328 static inline struct kmem_cache_order_objects oo_make(unsigned int order,
329 		unsigned int size)
330 {
331 	struct kmem_cache_order_objects x = {
332 		(order << OO_SHIFT) + order_objects(order, size)
333 	};
334 
335 	return x;
336 }
337 
338 static inline unsigned int oo_order(struct kmem_cache_order_objects x)
339 {
340 	return x.x >> OO_SHIFT;
341 }
342 
343 static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
344 {
345 	return x.x & OO_MASK;
346 }
347 
348 /*
349  * Per slab locking using the pagelock
350  */
351 static __always_inline void slab_lock(struct page *page)
352 {
353 	VM_BUG_ON_PAGE(PageTail(page), page);
354 	bit_spin_lock(PG_locked, &page->flags);
355 }
356 
357 static __always_inline void slab_unlock(struct page *page)
358 {
359 	VM_BUG_ON_PAGE(PageTail(page), page);
360 	__bit_spin_unlock(PG_locked, &page->flags);
361 }
362 
363 /* Interrupts must be disabled (for the fallback code to work right) */
364 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
365 		void *freelist_old, unsigned long counters_old,
366 		void *freelist_new, unsigned long counters_new,
367 		const char *n)
368 {
369 	VM_BUG_ON(!irqs_disabled());
370 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
371     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
372 	if (s->flags & __CMPXCHG_DOUBLE) {
373 		if (cmpxchg_double(&page->freelist, &page->counters,
374 				   freelist_old, counters_old,
375 				   freelist_new, counters_new))
376 			return true;
377 	} else
378 #endif
379 	{
380 		slab_lock(page);
381 		if (page->freelist == freelist_old &&
382 					page->counters == counters_old) {
383 			page->freelist = freelist_new;
384 			page->counters = counters_new;
385 			slab_unlock(page);
386 			return true;
387 		}
388 		slab_unlock(page);
389 	}
390 
391 	cpu_relax();
392 	stat(s, CMPXCHG_DOUBLE_FAIL);
393 
394 #ifdef SLUB_DEBUG_CMPXCHG
395 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
396 #endif
397 
398 	return false;
399 }
400 
401 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
402 		void *freelist_old, unsigned long counters_old,
403 		void *freelist_new, unsigned long counters_new,
404 		const char *n)
405 {
406 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
407     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
408 	if (s->flags & __CMPXCHG_DOUBLE) {
409 		if (cmpxchg_double(&page->freelist, &page->counters,
410 				   freelist_old, counters_old,
411 				   freelist_new, counters_new))
412 			return true;
413 	} else
414 #endif
415 	{
416 		unsigned long flags;
417 
418 		local_irq_save(flags);
419 		slab_lock(page);
420 		if (page->freelist == freelist_old &&
421 					page->counters == counters_old) {
422 			page->freelist = freelist_new;
423 			page->counters = counters_new;
424 			slab_unlock(page);
425 			local_irq_restore(flags);
426 			return true;
427 		}
428 		slab_unlock(page);
429 		local_irq_restore(flags);
430 	}
431 
432 	cpu_relax();
433 	stat(s, CMPXCHG_DOUBLE_FAIL);
434 
435 #ifdef SLUB_DEBUG_CMPXCHG
436 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
437 #endif
438 
439 	return false;
440 }
441 
442 #ifdef CONFIG_SLUB_DEBUG
443 /*
444  * Determine a map of object in use on a page.
445  *
446  * Node listlock must be held to guarantee that the page does
447  * not vanish from under us.
448  */
449 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
450 {
451 	void *p;
452 	void *addr = page_address(page);
453 
454 	for (p = page->freelist; p; p = get_freepointer(s, p))
455 		set_bit(slab_index(p, s, addr), map);
456 }
457 
458 static inline unsigned int size_from_object(struct kmem_cache *s)
459 {
460 	if (s->flags & SLAB_RED_ZONE)
461 		return s->size - s->red_left_pad;
462 
463 	return s->size;
464 }
465 
466 static inline void *restore_red_left(struct kmem_cache *s, void *p)
467 {
468 	if (s->flags & SLAB_RED_ZONE)
469 		p -= s->red_left_pad;
470 
471 	return p;
472 }
473 
474 /*
475  * Debug settings:
476  */
477 #if defined(CONFIG_SLUB_DEBUG_ON)
478 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
479 #else
480 static slab_flags_t slub_debug;
481 #endif
482 
483 static char *slub_debug_slabs;
484 static int disable_higher_order_debug;
485 
486 /*
487  * slub is about to manipulate internal object metadata.  This memory lies
488  * outside the range of the allocated object, so accessing it would normally
489  * be reported by kasan as a bounds error.  metadata_access_enable() is used
490  * to tell kasan that these accesses are OK.
491  */
492 static inline void metadata_access_enable(void)
493 {
494 	kasan_disable_current();
495 }
496 
497 static inline void metadata_access_disable(void)
498 {
499 	kasan_enable_current();
500 }
501 
502 /*
503  * Object debugging
504  */
505 
506 /* Verify that a pointer has an address that is valid within a slab page */
507 static inline int check_valid_pointer(struct kmem_cache *s,
508 				struct page *page, void *object)
509 {
510 	void *base;
511 
512 	if (!object)
513 		return 1;
514 
515 	base = page_address(page);
516 	object = kasan_reset_tag(object);
517 	object = restore_red_left(s, object);
518 	if (object < base || object >= base + page->objects * s->size ||
519 		(object - base) % s->size) {
520 		return 0;
521 	}
522 
523 	return 1;
524 }
525 
526 static void print_section(char *level, char *text, u8 *addr,
527 			  unsigned int length)
528 {
529 	metadata_access_enable();
530 	print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
531 			length, 1);
532 	metadata_access_disable();
533 }
534 
535 static struct track *get_track(struct kmem_cache *s, void *object,
536 	enum track_item alloc)
537 {
538 	struct track *p;
539 
540 	if (s->offset)
541 		p = object + s->offset + sizeof(void *);
542 	else
543 		p = object + s->inuse;
544 
545 	return p + alloc;
546 }
547 
548 static void set_track(struct kmem_cache *s, void *object,
549 			enum track_item alloc, unsigned long addr)
550 {
551 	struct track *p = get_track(s, object, alloc);
552 
553 	if (addr) {
554 #ifdef CONFIG_STACKTRACE
555 		struct stack_trace trace;
556 		int i;
557 
558 		trace.nr_entries = 0;
559 		trace.max_entries = TRACK_ADDRS_COUNT;
560 		trace.entries = p->addrs;
561 		trace.skip = 3;
562 		metadata_access_enable();
563 		save_stack_trace(&trace);
564 		metadata_access_disable();
565 
566 		/* See rant in lockdep.c */
567 		if (trace.nr_entries != 0 &&
568 		    trace.entries[trace.nr_entries - 1] == ULONG_MAX)
569 			trace.nr_entries--;
570 
571 		for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
572 			p->addrs[i] = 0;
573 #endif
574 		p->addr = addr;
575 		p->cpu = smp_processor_id();
576 		p->pid = current->pid;
577 		p->when = jiffies;
578 	} else
579 		memset(p, 0, sizeof(struct track));
580 }
581 
582 static void init_tracking(struct kmem_cache *s, void *object)
583 {
584 	if (!(s->flags & SLAB_STORE_USER))
585 		return;
586 
587 	set_track(s, object, TRACK_FREE, 0UL);
588 	set_track(s, object, TRACK_ALLOC, 0UL);
589 }
590 
591 static void print_track(const char *s, struct track *t, unsigned long pr_time)
592 {
593 	if (!t->addr)
594 		return;
595 
596 	pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
597 	       s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
598 #ifdef CONFIG_STACKTRACE
599 	{
600 		int i;
601 		for (i = 0; i < TRACK_ADDRS_COUNT; i++)
602 			if (t->addrs[i])
603 				pr_err("\t%pS\n", (void *)t->addrs[i]);
604 			else
605 				break;
606 	}
607 #endif
608 }
609 
610 static void print_tracking(struct kmem_cache *s, void *object)
611 {
612 	unsigned long pr_time = jiffies;
613 	if (!(s->flags & SLAB_STORE_USER))
614 		return;
615 
616 	print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
617 	print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
618 }
619 
620 static void print_page_info(struct page *page)
621 {
622 	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
623 	       page, page->objects, page->inuse, page->freelist, page->flags);
624 
625 }
626 
627 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
628 {
629 	struct va_format vaf;
630 	va_list args;
631 
632 	va_start(args, fmt);
633 	vaf.fmt = fmt;
634 	vaf.va = &args;
635 	pr_err("=============================================================================\n");
636 	pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
637 	pr_err("-----------------------------------------------------------------------------\n\n");
638 
639 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
640 	va_end(args);
641 }
642 
643 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
644 {
645 	struct va_format vaf;
646 	va_list args;
647 
648 	va_start(args, fmt);
649 	vaf.fmt = fmt;
650 	vaf.va = &args;
651 	pr_err("FIX %s: %pV\n", s->name, &vaf);
652 	va_end(args);
653 }
654 
655 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
656 {
657 	unsigned int off;	/* Offset of last byte */
658 	u8 *addr = page_address(page);
659 
660 	print_tracking(s, p);
661 
662 	print_page_info(page);
663 
664 	pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
665 	       p, p - addr, get_freepointer(s, p));
666 
667 	if (s->flags & SLAB_RED_ZONE)
668 		print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
669 			      s->red_left_pad);
670 	else if (p > addr + 16)
671 		print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
672 
673 	print_section(KERN_ERR, "Object ", p,
674 		      min_t(unsigned int, s->object_size, PAGE_SIZE));
675 	if (s->flags & SLAB_RED_ZONE)
676 		print_section(KERN_ERR, "Redzone ", p + s->object_size,
677 			s->inuse - s->object_size);
678 
679 	if (s->offset)
680 		off = s->offset + sizeof(void *);
681 	else
682 		off = s->inuse;
683 
684 	if (s->flags & SLAB_STORE_USER)
685 		off += 2 * sizeof(struct track);
686 
687 	off += kasan_metadata_size(s);
688 
689 	if (off != size_from_object(s))
690 		/* Beginning of the filler is the free pointer */
691 		print_section(KERN_ERR, "Padding ", p + off,
692 			      size_from_object(s) - off);
693 
694 	dump_stack();
695 }
696 
697 void object_err(struct kmem_cache *s, struct page *page,
698 			u8 *object, char *reason)
699 {
700 	slab_bug(s, "%s", reason);
701 	print_trailer(s, page, object);
702 }
703 
704 static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
705 			const char *fmt, ...)
706 {
707 	va_list args;
708 	char buf[100];
709 
710 	va_start(args, fmt);
711 	vsnprintf(buf, sizeof(buf), fmt, args);
712 	va_end(args);
713 	slab_bug(s, "%s", buf);
714 	print_page_info(page);
715 	dump_stack();
716 }
717 
718 static void init_object(struct kmem_cache *s, void *object, u8 val)
719 {
720 	u8 *p = object;
721 
722 	if (s->flags & SLAB_RED_ZONE)
723 		memset(p - s->red_left_pad, val, s->red_left_pad);
724 
725 	if (s->flags & __OBJECT_POISON) {
726 		memset(p, POISON_FREE, s->object_size - 1);
727 		p[s->object_size - 1] = POISON_END;
728 	}
729 
730 	if (s->flags & SLAB_RED_ZONE)
731 		memset(p + s->object_size, val, s->inuse - s->object_size);
732 }
733 
734 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
735 						void *from, void *to)
736 {
737 	slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
738 	memset(from, data, to - from);
739 }
740 
741 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
742 			u8 *object, char *what,
743 			u8 *start, unsigned int value, unsigned int bytes)
744 {
745 	u8 *fault;
746 	u8 *end;
747 
748 	metadata_access_enable();
749 	fault = memchr_inv(start, value, bytes);
750 	metadata_access_disable();
751 	if (!fault)
752 		return 1;
753 
754 	end = start + bytes;
755 	while (end > fault && end[-1] == value)
756 		end--;
757 
758 	slab_bug(s, "%s overwritten", what);
759 	pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
760 					fault, end - 1, fault[0], value);
761 	print_trailer(s, page, object);
762 
763 	restore_bytes(s, what, value, fault, end);
764 	return 0;
765 }
766 
767 /*
768  * Object layout:
769  *
770  * object address
771  * 	Bytes of the object to be managed.
772  * 	If the freepointer may overlay the object then the free
773  * 	pointer is the first word of the object.
774  *
775  * 	Poisoning uses 0x6b (POISON_FREE) and the last byte is
776  * 	0xa5 (POISON_END)
777  *
778  * object + s->object_size
779  * 	Padding to reach word boundary. This is also used for Redzoning.
780  * 	Padding is extended by another word if Redzoning is enabled and
781  * 	object_size == inuse.
782  *
783  * 	We fill with 0xbb (RED_INACTIVE) for inactive objects and with
784  * 	0xcc (RED_ACTIVE) for objects in use.
785  *
786  * object + s->inuse
787  * 	Meta data starts here.
788  *
789  * 	A. Free pointer (if we cannot overwrite object on free)
790  * 	B. Tracking data for SLAB_STORE_USER
791  * 	C. Padding to reach required alignment boundary or at mininum
792  * 		one word if debugging is on to be able to detect writes
793  * 		before the word boundary.
794  *
795  *	Padding is done using 0x5a (POISON_INUSE)
796  *
797  * object + s->size
798  * 	Nothing is used beyond s->size.
799  *
800  * If slabcaches are merged then the object_size and inuse boundaries are mostly
801  * ignored. And therefore no slab options that rely on these boundaries
802  * may be used with merged slabcaches.
803  */
804 
805 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
806 {
807 	unsigned long off = s->inuse;	/* The end of info */
808 
809 	if (s->offset)
810 		/* Freepointer is placed after the object. */
811 		off += sizeof(void *);
812 
813 	if (s->flags & SLAB_STORE_USER)
814 		/* We also have user information there */
815 		off += 2 * sizeof(struct track);
816 
817 	off += kasan_metadata_size(s);
818 
819 	if (size_from_object(s) == off)
820 		return 1;
821 
822 	return check_bytes_and_report(s, page, p, "Object padding",
823 			p + off, POISON_INUSE, size_from_object(s) - off);
824 }
825 
826 /* Check the pad bytes at the end of a slab page */
827 static int slab_pad_check(struct kmem_cache *s, struct page *page)
828 {
829 	u8 *start;
830 	u8 *fault;
831 	u8 *end;
832 	u8 *pad;
833 	int length;
834 	int remainder;
835 
836 	if (!(s->flags & SLAB_POISON))
837 		return 1;
838 
839 	start = page_address(page);
840 	length = PAGE_SIZE << compound_order(page);
841 	end = start + length;
842 	remainder = length % s->size;
843 	if (!remainder)
844 		return 1;
845 
846 	pad = end - remainder;
847 	metadata_access_enable();
848 	fault = memchr_inv(pad, POISON_INUSE, remainder);
849 	metadata_access_disable();
850 	if (!fault)
851 		return 1;
852 	while (end > fault && end[-1] == POISON_INUSE)
853 		end--;
854 
855 	slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
856 	print_section(KERN_ERR, "Padding ", pad, remainder);
857 
858 	restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
859 	return 0;
860 }
861 
862 static int check_object(struct kmem_cache *s, struct page *page,
863 					void *object, u8 val)
864 {
865 	u8 *p = object;
866 	u8 *endobject = object + s->object_size;
867 
868 	if (s->flags & SLAB_RED_ZONE) {
869 		if (!check_bytes_and_report(s, page, object, "Redzone",
870 			object - s->red_left_pad, val, s->red_left_pad))
871 			return 0;
872 
873 		if (!check_bytes_and_report(s, page, object, "Redzone",
874 			endobject, val, s->inuse - s->object_size))
875 			return 0;
876 	} else {
877 		if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
878 			check_bytes_and_report(s, page, p, "Alignment padding",
879 				endobject, POISON_INUSE,
880 				s->inuse - s->object_size);
881 		}
882 	}
883 
884 	if (s->flags & SLAB_POISON) {
885 		if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
886 			(!check_bytes_and_report(s, page, p, "Poison", p,
887 					POISON_FREE, s->object_size - 1) ||
888 			 !check_bytes_and_report(s, page, p, "Poison",
889 				p + s->object_size - 1, POISON_END, 1)))
890 			return 0;
891 		/*
892 		 * check_pad_bytes cleans up on its own.
893 		 */
894 		check_pad_bytes(s, page, p);
895 	}
896 
897 	if (!s->offset && val == SLUB_RED_ACTIVE)
898 		/*
899 		 * Object and freepointer overlap. Cannot check
900 		 * freepointer while object is allocated.
901 		 */
902 		return 1;
903 
904 	/* Check free pointer validity */
905 	if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
906 		object_err(s, page, p, "Freepointer corrupt");
907 		/*
908 		 * No choice but to zap it and thus lose the remainder
909 		 * of the free objects in this slab. May cause
910 		 * another error because the object count is now wrong.
911 		 */
912 		set_freepointer(s, p, NULL);
913 		return 0;
914 	}
915 	return 1;
916 }
917 
918 static int check_slab(struct kmem_cache *s, struct page *page)
919 {
920 	int maxobj;
921 
922 	VM_BUG_ON(!irqs_disabled());
923 
924 	if (!PageSlab(page)) {
925 		slab_err(s, page, "Not a valid slab page");
926 		return 0;
927 	}
928 
929 	maxobj = order_objects(compound_order(page), s->size);
930 	if (page->objects > maxobj) {
931 		slab_err(s, page, "objects %u > max %u",
932 			page->objects, maxobj);
933 		return 0;
934 	}
935 	if (page->inuse > page->objects) {
936 		slab_err(s, page, "inuse %u > max %u",
937 			page->inuse, page->objects);
938 		return 0;
939 	}
940 	/* Slab_pad_check fixes things up after itself */
941 	slab_pad_check(s, page);
942 	return 1;
943 }
944 
945 /*
946  * Determine if a certain object on a page is on the freelist. Must hold the
947  * slab lock to guarantee that the chains are in a consistent state.
948  */
949 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
950 {
951 	int nr = 0;
952 	void *fp;
953 	void *object = NULL;
954 	int max_objects;
955 
956 	fp = page->freelist;
957 	while (fp && nr <= page->objects) {
958 		if (fp == search)
959 			return 1;
960 		if (!check_valid_pointer(s, page, fp)) {
961 			if (object) {
962 				object_err(s, page, object,
963 					"Freechain corrupt");
964 				set_freepointer(s, object, NULL);
965 			} else {
966 				slab_err(s, page, "Freepointer corrupt");
967 				page->freelist = NULL;
968 				page->inuse = page->objects;
969 				slab_fix(s, "Freelist cleared");
970 				return 0;
971 			}
972 			break;
973 		}
974 		object = fp;
975 		fp = get_freepointer(s, object);
976 		nr++;
977 	}
978 
979 	max_objects = order_objects(compound_order(page), s->size);
980 	if (max_objects > MAX_OBJS_PER_PAGE)
981 		max_objects = MAX_OBJS_PER_PAGE;
982 
983 	if (page->objects != max_objects) {
984 		slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
985 			 page->objects, max_objects);
986 		page->objects = max_objects;
987 		slab_fix(s, "Number of objects adjusted.");
988 	}
989 	if (page->inuse != page->objects - nr) {
990 		slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
991 			 page->inuse, page->objects - nr);
992 		page->inuse = page->objects - nr;
993 		slab_fix(s, "Object count adjusted.");
994 	}
995 	return search == NULL;
996 }
997 
998 static void trace(struct kmem_cache *s, struct page *page, void *object,
999 								int alloc)
1000 {
1001 	if (s->flags & SLAB_TRACE) {
1002 		pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1003 			s->name,
1004 			alloc ? "alloc" : "free",
1005 			object, page->inuse,
1006 			page->freelist);
1007 
1008 		if (!alloc)
1009 			print_section(KERN_INFO, "Object ", (void *)object,
1010 					s->object_size);
1011 
1012 		dump_stack();
1013 	}
1014 }
1015 
1016 /*
1017  * Tracking of fully allocated slabs for debugging purposes.
1018  */
1019 static void add_full(struct kmem_cache *s,
1020 	struct kmem_cache_node *n, struct page *page)
1021 {
1022 	if (!(s->flags & SLAB_STORE_USER))
1023 		return;
1024 
1025 	lockdep_assert_held(&n->list_lock);
1026 	list_add(&page->lru, &n->full);
1027 }
1028 
1029 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
1030 {
1031 	if (!(s->flags & SLAB_STORE_USER))
1032 		return;
1033 
1034 	lockdep_assert_held(&n->list_lock);
1035 	list_del(&page->lru);
1036 }
1037 
1038 /* Tracking of the number of slabs for debugging purposes */
1039 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1040 {
1041 	struct kmem_cache_node *n = get_node(s, node);
1042 
1043 	return atomic_long_read(&n->nr_slabs);
1044 }
1045 
1046 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1047 {
1048 	return atomic_long_read(&n->nr_slabs);
1049 }
1050 
1051 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1052 {
1053 	struct kmem_cache_node *n = get_node(s, node);
1054 
1055 	/*
1056 	 * May be called early in order to allocate a slab for the
1057 	 * kmem_cache_node structure. Solve the chicken-egg
1058 	 * dilemma by deferring the increment of the count during
1059 	 * bootstrap (see early_kmem_cache_node_alloc).
1060 	 */
1061 	if (likely(n)) {
1062 		atomic_long_inc(&n->nr_slabs);
1063 		atomic_long_add(objects, &n->total_objects);
1064 	}
1065 }
1066 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1067 {
1068 	struct kmem_cache_node *n = get_node(s, node);
1069 
1070 	atomic_long_dec(&n->nr_slabs);
1071 	atomic_long_sub(objects, &n->total_objects);
1072 }
1073 
1074 /* Object debug checks for alloc/free paths */
1075 static void setup_object_debug(struct kmem_cache *s, struct page *page,
1076 								void *object)
1077 {
1078 	if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1079 		return;
1080 
1081 	init_object(s, object, SLUB_RED_INACTIVE);
1082 	init_tracking(s, object);
1083 }
1084 
1085 static void setup_page_debug(struct kmem_cache *s, void *addr, int order)
1086 {
1087 	if (!(s->flags & SLAB_POISON))
1088 		return;
1089 
1090 	metadata_access_enable();
1091 	memset(addr, POISON_INUSE, PAGE_SIZE << order);
1092 	metadata_access_disable();
1093 }
1094 
1095 static inline int alloc_consistency_checks(struct kmem_cache *s,
1096 					struct page *page, void *object)
1097 {
1098 	if (!check_slab(s, page))
1099 		return 0;
1100 
1101 	if (!check_valid_pointer(s, page, object)) {
1102 		object_err(s, page, object, "Freelist Pointer check fails");
1103 		return 0;
1104 	}
1105 
1106 	if (!check_object(s, page, object, SLUB_RED_INACTIVE))
1107 		return 0;
1108 
1109 	return 1;
1110 }
1111 
1112 static noinline int alloc_debug_processing(struct kmem_cache *s,
1113 					struct page *page,
1114 					void *object, unsigned long addr)
1115 {
1116 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1117 		if (!alloc_consistency_checks(s, page, object))
1118 			goto bad;
1119 	}
1120 
1121 	/* Success perform special debug activities for allocs */
1122 	if (s->flags & SLAB_STORE_USER)
1123 		set_track(s, object, TRACK_ALLOC, addr);
1124 	trace(s, page, object, 1);
1125 	init_object(s, object, SLUB_RED_ACTIVE);
1126 	return 1;
1127 
1128 bad:
1129 	if (PageSlab(page)) {
1130 		/*
1131 		 * If this is a slab page then lets do the best we can
1132 		 * to avoid issues in the future. Marking all objects
1133 		 * as used avoids touching the remaining objects.
1134 		 */
1135 		slab_fix(s, "Marking all objects used");
1136 		page->inuse = page->objects;
1137 		page->freelist = NULL;
1138 	}
1139 	return 0;
1140 }
1141 
1142 static inline int free_consistency_checks(struct kmem_cache *s,
1143 		struct page *page, void *object, unsigned long addr)
1144 {
1145 	if (!check_valid_pointer(s, page, object)) {
1146 		slab_err(s, page, "Invalid object pointer 0x%p", object);
1147 		return 0;
1148 	}
1149 
1150 	if (on_freelist(s, page, object)) {
1151 		object_err(s, page, object, "Object already free");
1152 		return 0;
1153 	}
1154 
1155 	if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1156 		return 0;
1157 
1158 	if (unlikely(s != page->slab_cache)) {
1159 		if (!PageSlab(page)) {
1160 			slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1161 				 object);
1162 		} else if (!page->slab_cache) {
1163 			pr_err("SLUB <none>: no slab for object 0x%p.\n",
1164 			       object);
1165 			dump_stack();
1166 		} else
1167 			object_err(s, page, object,
1168 					"page slab pointer corrupt.");
1169 		return 0;
1170 	}
1171 	return 1;
1172 }
1173 
1174 /* Supports checking bulk free of a constructed freelist */
1175 static noinline int free_debug_processing(
1176 	struct kmem_cache *s, struct page *page,
1177 	void *head, void *tail, int bulk_cnt,
1178 	unsigned long addr)
1179 {
1180 	struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1181 	void *object = head;
1182 	int cnt = 0;
1183 	unsigned long uninitialized_var(flags);
1184 	int ret = 0;
1185 
1186 	spin_lock_irqsave(&n->list_lock, flags);
1187 	slab_lock(page);
1188 
1189 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1190 		if (!check_slab(s, page))
1191 			goto out;
1192 	}
1193 
1194 next_object:
1195 	cnt++;
1196 
1197 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1198 		if (!free_consistency_checks(s, page, object, addr))
1199 			goto out;
1200 	}
1201 
1202 	if (s->flags & SLAB_STORE_USER)
1203 		set_track(s, object, TRACK_FREE, addr);
1204 	trace(s, page, object, 0);
1205 	/* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1206 	init_object(s, object, SLUB_RED_INACTIVE);
1207 
1208 	/* Reached end of constructed freelist yet? */
1209 	if (object != tail) {
1210 		object = get_freepointer(s, object);
1211 		goto next_object;
1212 	}
1213 	ret = 1;
1214 
1215 out:
1216 	if (cnt != bulk_cnt)
1217 		slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1218 			 bulk_cnt, cnt);
1219 
1220 	slab_unlock(page);
1221 	spin_unlock_irqrestore(&n->list_lock, flags);
1222 	if (!ret)
1223 		slab_fix(s, "Object at 0x%p not freed", object);
1224 	return ret;
1225 }
1226 
1227 static int __init setup_slub_debug(char *str)
1228 {
1229 	slub_debug = DEBUG_DEFAULT_FLAGS;
1230 	if (*str++ != '=' || !*str)
1231 		/*
1232 		 * No options specified. Switch on full debugging.
1233 		 */
1234 		goto out;
1235 
1236 	if (*str == ',')
1237 		/*
1238 		 * No options but restriction on slabs. This means full
1239 		 * debugging for slabs matching a pattern.
1240 		 */
1241 		goto check_slabs;
1242 
1243 	slub_debug = 0;
1244 	if (*str == '-')
1245 		/*
1246 		 * Switch off all debugging measures.
1247 		 */
1248 		goto out;
1249 
1250 	/*
1251 	 * Determine which debug features should be switched on
1252 	 */
1253 	for (; *str && *str != ','; str++) {
1254 		switch (tolower(*str)) {
1255 		case 'f':
1256 			slub_debug |= SLAB_CONSISTENCY_CHECKS;
1257 			break;
1258 		case 'z':
1259 			slub_debug |= SLAB_RED_ZONE;
1260 			break;
1261 		case 'p':
1262 			slub_debug |= SLAB_POISON;
1263 			break;
1264 		case 'u':
1265 			slub_debug |= SLAB_STORE_USER;
1266 			break;
1267 		case 't':
1268 			slub_debug |= SLAB_TRACE;
1269 			break;
1270 		case 'a':
1271 			slub_debug |= SLAB_FAILSLAB;
1272 			break;
1273 		case 'o':
1274 			/*
1275 			 * Avoid enabling debugging on caches if its minimum
1276 			 * order would increase as a result.
1277 			 */
1278 			disable_higher_order_debug = 1;
1279 			break;
1280 		default:
1281 			pr_err("slub_debug option '%c' unknown. skipped\n",
1282 			       *str);
1283 		}
1284 	}
1285 
1286 check_slabs:
1287 	if (*str == ',')
1288 		slub_debug_slabs = str + 1;
1289 out:
1290 	return 1;
1291 }
1292 
1293 __setup("slub_debug", setup_slub_debug);
1294 
1295 /*
1296  * kmem_cache_flags - apply debugging options to the cache
1297  * @object_size:	the size of an object without meta data
1298  * @flags:		flags to set
1299  * @name:		name of the cache
1300  * @ctor:		constructor function
1301  *
1302  * Debug option(s) are applied to @flags. In addition to the debug
1303  * option(s), if a slab name (or multiple) is specified i.e.
1304  * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1305  * then only the select slabs will receive the debug option(s).
1306  */
1307 slab_flags_t kmem_cache_flags(unsigned int object_size,
1308 	slab_flags_t flags, const char *name,
1309 	void (*ctor)(void *))
1310 {
1311 	char *iter;
1312 	size_t len;
1313 
1314 	/* If slub_debug = 0, it folds into the if conditional. */
1315 	if (!slub_debug_slabs)
1316 		return flags | slub_debug;
1317 
1318 	len = strlen(name);
1319 	iter = slub_debug_slabs;
1320 	while (*iter) {
1321 		char *end, *glob;
1322 		size_t cmplen;
1323 
1324 		end = strchr(iter, ',');
1325 		if (!end)
1326 			end = iter + strlen(iter);
1327 
1328 		glob = strnchr(iter, end - iter, '*');
1329 		if (glob)
1330 			cmplen = glob - iter;
1331 		else
1332 			cmplen = max_t(size_t, len, (end - iter));
1333 
1334 		if (!strncmp(name, iter, cmplen)) {
1335 			flags |= slub_debug;
1336 			break;
1337 		}
1338 
1339 		if (!*end)
1340 			break;
1341 		iter = end + 1;
1342 	}
1343 
1344 	return flags;
1345 }
1346 #else /* !CONFIG_SLUB_DEBUG */
1347 static inline void setup_object_debug(struct kmem_cache *s,
1348 			struct page *page, void *object) {}
1349 static inline void setup_page_debug(struct kmem_cache *s,
1350 			void *addr, int order) {}
1351 
1352 static inline int alloc_debug_processing(struct kmem_cache *s,
1353 	struct page *page, void *object, unsigned long addr) { return 0; }
1354 
1355 static inline int free_debug_processing(
1356 	struct kmem_cache *s, struct page *page,
1357 	void *head, void *tail, int bulk_cnt,
1358 	unsigned long addr) { return 0; }
1359 
1360 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1361 			{ return 1; }
1362 static inline int check_object(struct kmem_cache *s, struct page *page,
1363 			void *object, u8 val) { return 1; }
1364 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1365 					struct page *page) {}
1366 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1367 					struct page *page) {}
1368 slab_flags_t kmem_cache_flags(unsigned int object_size,
1369 	slab_flags_t flags, const char *name,
1370 	void (*ctor)(void *))
1371 {
1372 	return flags;
1373 }
1374 #define slub_debug 0
1375 
1376 #define disable_higher_order_debug 0
1377 
1378 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1379 							{ return 0; }
1380 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1381 							{ return 0; }
1382 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1383 							int objects) {}
1384 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1385 							int objects) {}
1386 
1387 #endif /* CONFIG_SLUB_DEBUG */
1388 
1389 /*
1390  * Hooks for other subsystems that check memory allocations. In a typical
1391  * production configuration these hooks all should produce no code at all.
1392  */
1393 static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1394 {
1395 	ptr = kasan_kmalloc_large(ptr, size, flags);
1396 	/* As ptr might get tagged, call kmemleak hook after KASAN. */
1397 	kmemleak_alloc(ptr, size, 1, flags);
1398 	return ptr;
1399 }
1400 
1401 static __always_inline void kfree_hook(void *x)
1402 {
1403 	kmemleak_free(x);
1404 	kasan_kfree_large(x, _RET_IP_);
1405 }
1406 
1407 static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x)
1408 {
1409 	kmemleak_free_recursive(x, s->flags);
1410 
1411 	/*
1412 	 * Trouble is that we may no longer disable interrupts in the fast path
1413 	 * So in order to make the debug calls that expect irqs to be
1414 	 * disabled we need to disable interrupts temporarily.
1415 	 */
1416 #ifdef CONFIG_LOCKDEP
1417 	{
1418 		unsigned long flags;
1419 
1420 		local_irq_save(flags);
1421 		debug_check_no_locks_freed(x, s->object_size);
1422 		local_irq_restore(flags);
1423 	}
1424 #endif
1425 	if (!(s->flags & SLAB_DEBUG_OBJECTS))
1426 		debug_check_no_obj_freed(x, s->object_size);
1427 
1428 	/* KASAN might put x into memory quarantine, delaying its reuse */
1429 	return kasan_slab_free(s, x, _RET_IP_);
1430 }
1431 
1432 static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1433 					   void **head, void **tail)
1434 {
1435 /*
1436  * Compiler cannot detect this function can be removed if slab_free_hook()
1437  * evaluates to nothing.  Thus, catch all relevant config debug options here.
1438  */
1439 #if defined(CONFIG_LOCKDEP)	||		\
1440 	defined(CONFIG_DEBUG_KMEMLEAK) ||	\
1441 	defined(CONFIG_DEBUG_OBJECTS_FREE) ||	\
1442 	defined(CONFIG_KASAN)
1443 
1444 	void *object;
1445 	void *next = *head;
1446 	void *old_tail = *tail ? *tail : *head;
1447 
1448 	/* Head and tail of the reconstructed freelist */
1449 	*head = NULL;
1450 	*tail = NULL;
1451 
1452 	do {
1453 		object = next;
1454 		next = get_freepointer(s, object);
1455 		/* If object's reuse doesn't have to be delayed */
1456 		if (!slab_free_hook(s, object)) {
1457 			/* Move object to the new freelist */
1458 			set_freepointer(s, object, *head);
1459 			*head = object;
1460 			if (!*tail)
1461 				*tail = object;
1462 		}
1463 	} while (object != old_tail);
1464 
1465 	if (*head == *tail)
1466 		*tail = NULL;
1467 
1468 	return *head != NULL;
1469 #else
1470 	return true;
1471 #endif
1472 }
1473 
1474 static void *setup_object(struct kmem_cache *s, struct page *page,
1475 				void *object)
1476 {
1477 	setup_object_debug(s, page, object);
1478 	object = kasan_init_slab_obj(s, object);
1479 	if (unlikely(s->ctor)) {
1480 		kasan_unpoison_object_data(s, object);
1481 		s->ctor(object);
1482 		kasan_poison_object_data(s, object);
1483 	}
1484 	return object;
1485 }
1486 
1487 /*
1488  * Slab allocation and freeing
1489  */
1490 static inline struct page *alloc_slab_page(struct kmem_cache *s,
1491 		gfp_t flags, int node, struct kmem_cache_order_objects oo)
1492 {
1493 	struct page *page;
1494 	unsigned int order = oo_order(oo);
1495 
1496 	if (node == NUMA_NO_NODE)
1497 		page = alloc_pages(flags, order);
1498 	else
1499 		page = __alloc_pages_node(node, flags, order);
1500 
1501 	if (page && memcg_charge_slab(page, flags, order, s)) {
1502 		__free_pages(page, order);
1503 		page = NULL;
1504 	}
1505 
1506 	return page;
1507 }
1508 
1509 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1510 /* Pre-initialize the random sequence cache */
1511 static int init_cache_random_seq(struct kmem_cache *s)
1512 {
1513 	unsigned int count = oo_objects(s->oo);
1514 	int err;
1515 
1516 	/* Bailout if already initialised */
1517 	if (s->random_seq)
1518 		return 0;
1519 
1520 	err = cache_random_seq_create(s, count, GFP_KERNEL);
1521 	if (err) {
1522 		pr_err("SLUB: Unable to initialize free list for %s\n",
1523 			s->name);
1524 		return err;
1525 	}
1526 
1527 	/* Transform to an offset on the set of pages */
1528 	if (s->random_seq) {
1529 		unsigned int i;
1530 
1531 		for (i = 0; i < count; i++)
1532 			s->random_seq[i] *= s->size;
1533 	}
1534 	return 0;
1535 }
1536 
1537 /* Initialize each random sequence freelist per cache */
1538 static void __init init_freelist_randomization(void)
1539 {
1540 	struct kmem_cache *s;
1541 
1542 	mutex_lock(&slab_mutex);
1543 
1544 	list_for_each_entry(s, &slab_caches, list)
1545 		init_cache_random_seq(s);
1546 
1547 	mutex_unlock(&slab_mutex);
1548 }
1549 
1550 /* Get the next entry on the pre-computed freelist randomized */
1551 static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1552 				unsigned long *pos, void *start,
1553 				unsigned long page_limit,
1554 				unsigned long freelist_count)
1555 {
1556 	unsigned int idx;
1557 
1558 	/*
1559 	 * If the target page allocation failed, the number of objects on the
1560 	 * page might be smaller than the usual size defined by the cache.
1561 	 */
1562 	do {
1563 		idx = s->random_seq[*pos];
1564 		*pos += 1;
1565 		if (*pos >= freelist_count)
1566 			*pos = 0;
1567 	} while (unlikely(idx >= page_limit));
1568 
1569 	return (char *)start + idx;
1570 }
1571 
1572 /* Shuffle the single linked freelist based on a random pre-computed sequence */
1573 static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1574 {
1575 	void *start;
1576 	void *cur;
1577 	void *next;
1578 	unsigned long idx, pos, page_limit, freelist_count;
1579 
1580 	if (page->objects < 2 || !s->random_seq)
1581 		return false;
1582 
1583 	freelist_count = oo_objects(s->oo);
1584 	pos = get_random_int() % freelist_count;
1585 
1586 	page_limit = page->objects * s->size;
1587 	start = fixup_red_left(s, page_address(page));
1588 
1589 	/* First entry is used as the base of the freelist */
1590 	cur = next_freelist_entry(s, page, &pos, start, page_limit,
1591 				freelist_count);
1592 	cur = setup_object(s, page, cur);
1593 	page->freelist = cur;
1594 
1595 	for (idx = 1; idx < page->objects; idx++) {
1596 		next = next_freelist_entry(s, page, &pos, start, page_limit,
1597 			freelist_count);
1598 		next = setup_object(s, page, next);
1599 		set_freepointer(s, cur, next);
1600 		cur = next;
1601 	}
1602 	set_freepointer(s, cur, NULL);
1603 
1604 	return true;
1605 }
1606 #else
1607 static inline int init_cache_random_seq(struct kmem_cache *s)
1608 {
1609 	return 0;
1610 }
1611 static inline void init_freelist_randomization(void) { }
1612 static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1613 {
1614 	return false;
1615 }
1616 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1617 
1618 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1619 {
1620 	struct page *page;
1621 	struct kmem_cache_order_objects oo = s->oo;
1622 	gfp_t alloc_gfp;
1623 	void *start, *p, *next;
1624 	int idx, order;
1625 	bool shuffle;
1626 
1627 	flags &= gfp_allowed_mask;
1628 
1629 	if (gfpflags_allow_blocking(flags))
1630 		local_irq_enable();
1631 
1632 	flags |= s->allocflags;
1633 
1634 	/*
1635 	 * Let the initial higher-order allocation fail under memory pressure
1636 	 * so we fall-back to the minimum order allocation.
1637 	 */
1638 	alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1639 	if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1640 		alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
1641 
1642 	page = alloc_slab_page(s, alloc_gfp, node, oo);
1643 	if (unlikely(!page)) {
1644 		oo = s->min;
1645 		alloc_gfp = flags;
1646 		/*
1647 		 * Allocation may have failed due to fragmentation.
1648 		 * Try a lower order alloc if possible
1649 		 */
1650 		page = alloc_slab_page(s, alloc_gfp, node, oo);
1651 		if (unlikely(!page))
1652 			goto out;
1653 		stat(s, ORDER_FALLBACK);
1654 	}
1655 
1656 	page->objects = oo_objects(oo);
1657 
1658 	order = compound_order(page);
1659 	page->slab_cache = s;
1660 	__SetPageSlab(page);
1661 	if (page_is_pfmemalloc(page))
1662 		SetPageSlabPfmemalloc(page);
1663 
1664 	kasan_poison_slab(page);
1665 
1666 	start = page_address(page);
1667 
1668 	setup_page_debug(s, start, order);
1669 
1670 	shuffle = shuffle_freelist(s, page);
1671 
1672 	if (!shuffle) {
1673 		start = fixup_red_left(s, start);
1674 		start = setup_object(s, page, start);
1675 		page->freelist = start;
1676 		for (idx = 0, p = start; idx < page->objects - 1; idx++) {
1677 			next = p + s->size;
1678 			next = setup_object(s, page, next);
1679 			set_freepointer(s, p, next);
1680 			p = next;
1681 		}
1682 		set_freepointer(s, p, NULL);
1683 	}
1684 
1685 	page->inuse = page->objects;
1686 	page->frozen = 1;
1687 
1688 out:
1689 	if (gfpflags_allow_blocking(flags))
1690 		local_irq_disable();
1691 	if (!page)
1692 		return NULL;
1693 
1694 	mod_lruvec_page_state(page,
1695 		(s->flags & SLAB_RECLAIM_ACCOUNT) ?
1696 		NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1697 		1 << oo_order(oo));
1698 
1699 	inc_slabs_node(s, page_to_nid(page), page->objects);
1700 
1701 	return page;
1702 }
1703 
1704 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1705 {
1706 	if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1707 		gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1708 		flags &= ~GFP_SLAB_BUG_MASK;
1709 		pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1710 				invalid_mask, &invalid_mask, flags, &flags);
1711 		dump_stack();
1712 	}
1713 
1714 	return allocate_slab(s,
1715 		flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1716 }
1717 
1718 static void __free_slab(struct kmem_cache *s, struct page *page)
1719 {
1720 	int order = compound_order(page);
1721 	int pages = 1 << order;
1722 
1723 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1724 		void *p;
1725 
1726 		slab_pad_check(s, page);
1727 		for_each_object(p, s, page_address(page),
1728 						page->objects)
1729 			check_object(s, page, p, SLUB_RED_INACTIVE);
1730 	}
1731 
1732 	mod_lruvec_page_state(page,
1733 		(s->flags & SLAB_RECLAIM_ACCOUNT) ?
1734 		NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1735 		-pages);
1736 
1737 	__ClearPageSlabPfmemalloc(page);
1738 	__ClearPageSlab(page);
1739 
1740 	page->mapping = NULL;
1741 	if (current->reclaim_state)
1742 		current->reclaim_state->reclaimed_slab += pages;
1743 	memcg_uncharge_slab(page, order, s);
1744 	__free_pages(page, order);
1745 }
1746 
1747 static void rcu_free_slab(struct rcu_head *h)
1748 {
1749 	struct page *page = container_of(h, struct page, rcu_head);
1750 
1751 	__free_slab(page->slab_cache, page);
1752 }
1753 
1754 static void free_slab(struct kmem_cache *s, struct page *page)
1755 {
1756 	if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
1757 		call_rcu(&page->rcu_head, rcu_free_slab);
1758 	} else
1759 		__free_slab(s, page);
1760 }
1761 
1762 static void discard_slab(struct kmem_cache *s, struct page *page)
1763 {
1764 	dec_slabs_node(s, page_to_nid(page), page->objects);
1765 	free_slab(s, page);
1766 }
1767 
1768 /*
1769  * Management of partially allocated slabs.
1770  */
1771 static inline void
1772 __add_partial(struct kmem_cache_node *n, struct page *page, int tail)
1773 {
1774 	n->nr_partial++;
1775 	if (tail == DEACTIVATE_TO_TAIL)
1776 		list_add_tail(&page->lru, &n->partial);
1777 	else
1778 		list_add(&page->lru, &n->partial);
1779 }
1780 
1781 static inline void add_partial(struct kmem_cache_node *n,
1782 				struct page *page, int tail)
1783 {
1784 	lockdep_assert_held(&n->list_lock);
1785 	__add_partial(n, page, tail);
1786 }
1787 
1788 static inline void remove_partial(struct kmem_cache_node *n,
1789 					struct page *page)
1790 {
1791 	lockdep_assert_held(&n->list_lock);
1792 	list_del(&page->lru);
1793 	n->nr_partial--;
1794 }
1795 
1796 /*
1797  * Remove slab from the partial list, freeze it and
1798  * return the pointer to the freelist.
1799  *
1800  * Returns a list of objects or NULL if it fails.
1801  */
1802 static inline void *acquire_slab(struct kmem_cache *s,
1803 		struct kmem_cache_node *n, struct page *page,
1804 		int mode, int *objects)
1805 {
1806 	void *freelist;
1807 	unsigned long counters;
1808 	struct page new;
1809 
1810 	lockdep_assert_held(&n->list_lock);
1811 
1812 	/*
1813 	 * Zap the freelist and set the frozen bit.
1814 	 * The old freelist is the list of objects for the
1815 	 * per cpu allocation list.
1816 	 */
1817 	freelist = page->freelist;
1818 	counters = page->counters;
1819 	new.counters = counters;
1820 	*objects = new.objects - new.inuse;
1821 	if (mode) {
1822 		new.inuse = page->objects;
1823 		new.freelist = NULL;
1824 	} else {
1825 		new.freelist = freelist;
1826 	}
1827 
1828 	VM_BUG_ON(new.frozen);
1829 	new.frozen = 1;
1830 
1831 	if (!__cmpxchg_double_slab(s, page,
1832 			freelist, counters,
1833 			new.freelist, new.counters,
1834 			"acquire_slab"))
1835 		return NULL;
1836 
1837 	remove_partial(n, page);
1838 	WARN_ON(!freelist);
1839 	return freelist;
1840 }
1841 
1842 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1843 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
1844 
1845 /*
1846  * Try to allocate a partial slab from a specific node.
1847  */
1848 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1849 				struct kmem_cache_cpu *c, gfp_t flags)
1850 {
1851 	struct page *page, *page2;
1852 	void *object = NULL;
1853 	unsigned int available = 0;
1854 	int objects;
1855 
1856 	/*
1857 	 * Racy check. If we mistakenly see no partial slabs then we
1858 	 * just allocate an empty slab. If we mistakenly try to get a
1859 	 * partial slab and there is none available then get_partials()
1860 	 * will return NULL.
1861 	 */
1862 	if (!n || !n->nr_partial)
1863 		return NULL;
1864 
1865 	spin_lock(&n->list_lock);
1866 	list_for_each_entry_safe(page, page2, &n->partial, lru) {
1867 		void *t;
1868 
1869 		if (!pfmemalloc_match(page, flags))
1870 			continue;
1871 
1872 		t = acquire_slab(s, n, page, object == NULL, &objects);
1873 		if (!t)
1874 			break;
1875 
1876 		available += objects;
1877 		if (!object) {
1878 			c->page = page;
1879 			stat(s, ALLOC_FROM_PARTIAL);
1880 			object = t;
1881 		} else {
1882 			put_cpu_partial(s, page, 0);
1883 			stat(s, CPU_PARTIAL_NODE);
1884 		}
1885 		if (!kmem_cache_has_cpu_partial(s)
1886 			|| available > slub_cpu_partial(s) / 2)
1887 			break;
1888 
1889 	}
1890 	spin_unlock(&n->list_lock);
1891 	return object;
1892 }
1893 
1894 /*
1895  * Get a page from somewhere. Search in increasing NUMA distances.
1896  */
1897 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
1898 		struct kmem_cache_cpu *c)
1899 {
1900 #ifdef CONFIG_NUMA
1901 	struct zonelist *zonelist;
1902 	struct zoneref *z;
1903 	struct zone *zone;
1904 	enum zone_type high_zoneidx = gfp_zone(flags);
1905 	void *object;
1906 	unsigned int cpuset_mems_cookie;
1907 
1908 	/*
1909 	 * The defrag ratio allows a configuration of the tradeoffs between
1910 	 * inter node defragmentation and node local allocations. A lower
1911 	 * defrag_ratio increases the tendency to do local allocations
1912 	 * instead of attempting to obtain partial slabs from other nodes.
1913 	 *
1914 	 * If the defrag_ratio is set to 0 then kmalloc() always
1915 	 * returns node local objects. If the ratio is higher then kmalloc()
1916 	 * may return off node objects because partial slabs are obtained
1917 	 * from other nodes and filled up.
1918 	 *
1919 	 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1920 	 * (which makes defrag_ratio = 1000) then every (well almost)
1921 	 * allocation will first attempt to defrag slab caches on other nodes.
1922 	 * This means scanning over all nodes to look for partial slabs which
1923 	 * may be expensive if we do it every time we are trying to find a slab
1924 	 * with available objects.
1925 	 */
1926 	if (!s->remote_node_defrag_ratio ||
1927 			get_cycles() % 1024 > s->remote_node_defrag_ratio)
1928 		return NULL;
1929 
1930 	do {
1931 		cpuset_mems_cookie = read_mems_allowed_begin();
1932 		zonelist = node_zonelist(mempolicy_slab_node(), flags);
1933 		for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1934 			struct kmem_cache_node *n;
1935 
1936 			n = get_node(s, zone_to_nid(zone));
1937 
1938 			if (n && cpuset_zone_allowed(zone, flags) &&
1939 					n->nr_partial > s->min_partial) {
1940 				object = get_partial_node(s, n, c, flags);
1941 				if (object) {
1942 					/*
1943 					 * Don't check read_mems_allowed_retry()
1944 					 * here - if mems_allowed was updated in
1945 					 * parallel, that was a harmless race
1946 					 * between allocation and the cpuset
1947 					 * update
1948 					 */
1949 					return object;
1950 				}
1951 			}
1952 		}
1953 	} while (read_mems_allowed_retry(cpuset_mems_cookie));
1954 #endif
1955 	return NULL;
1956 }
1957 
1958 /*
1959  * Get a partial page, lock it and return it.
1960  */
1961 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
1962 		struct kmem_cache_cpu *c)
1963 {
1964 	void *object;
1965 	int searchnode = node;
1966 
1967 	if (node == NUMA_NO_NODE)
1968 		searchnode = numa_mem_id();
1969 	else if (!node_present_pages(node))
1970 		searchnode = node_to_mem_node(node);
1971 
1972 	object = get_partial_node(s, get_node(s, searchnode), c, flags);
1973 	if (object || node != NUMA_NO_NODE)
1974 		return object;
1975 
1976 	return get_any_partial(s, flags, c);
1977 }
1978 
1979 #ifdef CONFIG_PREEMPT
1980 /*
1981  * Calculate the next globally unique transaction for disambiguiation
1982  * during cmpxchg. The transactions start with the cpu number and are then
1983  * incremented by CONFIG_NR_CPUS.
1984  */
1985 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
1986 #else
1987 /*
1988  * No preemption supported therefore also no need to check for
1989  * different cpus.
1990  */
1991 #define TID_STEP 1
1992 #endif
1993 
1994 static inline unsigned long next_tid(unsigned long tid)
1995 {
1996 	return tid + TID_STEP;
1997 }
1998 
1999 static inline unsigned int tid_to_cpu(unsigned long tid)
2000 {
2001 	return tid % TID_STEP;
2002 }
2003 
2004 static inline unsigned long tid_to_event(unsigned long tid)
2005 {
2006 	return tid / TID_STEP;
2007 }
2008 
2009 static inline unsigned int init_tid(int cpu)
2010 {
2011 	return cpu;
2012 }
2013 
2014 static inline void note_cmpxchg_failure(const char *n,
2015 		const struct kmem_cache *s, unsigned long tid)
2016 {
2017 #ifdef SLUB_DEBUG_CMPXCHG
2018 	unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2019 
2020 	pr_info("%s %s: cmpxchg redo ", n, s->name);
2021 
2022 #ifdef CONFIG_PREEMPT
2023 	if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2024 		pr_warn("due to cpu change %d -> %d\n",
2025 			tid_to_cpu(tid), tid_to_cpu(actual_tid));
2026 	else
2027 #endif
2028 	if (tid_to_event(tid) != tid_to_event(actual_tid))
2029 		pr_warn("due to cpu running other code. Event %ld->%ld\n",
2030 			tid_to_event(tid), tid_to_event(actual_tid));
2031 	else
2032 		pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2033 			actual_tid, tid, next_tid(tid));
2034 #endif
2035 	stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2036 }
2037 
2038 static void init_kmem_cache_cpus(struct kmem_cache *s)
2039 {
2040 	int cpu;
2041 
2042 	for_each_possible_cpu(cpu)
2043 		per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
2044 }
2045 
2046 /*
2047  * Remove the cpu slab
2048  */
2049 static void deactivate_slab(struct kmem_cache *s, struct page *page,
2050 				void *freelist, struct kmem_cache_cpu *c)
2051 {
2052 	enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
2053 	struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2054 	int lock = 0;
2055 	enum slab_modes l = M_NONE, m = M_NONE;
2056 	void *nextfree;
2057 	int tail = DEACTIVATE_TO_HEAD;
2058 	struct page new;
2059 	struct page old;
2060 
2061 	if (page->freelist) {
2062 		stat(s, DEACTIVATE_REMOTE_FREES);
2063 		tail = DEACTIVATE_TO_TAIL;
2064 	}
2065 
2066 	/*
2067 	 * Stage one: Free all available per cpu objects back
2068 	 * to the page freelist while it is still frozen. Leave the
2069 	 * last one.
2070 	 *
2071 	 * There is no need to take the list->lock because the page
2072 	 * is still frozen.
2073 	 */
2074 	while (freelist && (nextfree = get_freepointer(s, freelist))) {
2075 		void *prior;
2076 		unsigned long counters;
2077 
2078 		do {
2079 			prior = page->freelist;
2080 			counters = page->counters;
2081 			set_freepointer(s, freelist, prior);
2082 			new.counters = counters;
2083 			new.inuse--;
2084 			VM_BUG_ON(!new.frozen);
2085 
2086 		} while (!__cmpxchg_double_slab(s, page,
2087 			prior, counters,
2088 			freelist, new.counters,
2089 			"drain percpu freelist"));
2090 
2091 		freelist = nextfree;
2092 	}
2093 
2094 	/*
2095 	 * Stage two: Ensure that the page is unfrozen while the
2096 	 * list presence reflects the actual number of objects
2097 	 * during unfreeze.
2098 	 *
2099 	 * We setup the list membership and then perform a cmpxchg
2100 	 * with the count. If there is a mismatch then the page
2101 	 * is not unfrozen but the page is on the wrong list.
2102 	 *
2103 	 * Then we restart the process which may have to remove
2104 	 * the page from the list that we just put it on again
2105 	 * because the number of objects in the slab may have
2106 	 * changed.
2107 	 */
2108 redo:
2109 
2110 	old.freelist = page->freelist;
2111 	old.counters = page->counters;
2112 	VM_BUG_ON(!old.frozen);
2113 
2114 	/* Determine target state of the slab */
2115 	new.counters = old.counters;
2116 	if (freelist) {
2117 		new.inuse--;
2118 		set_freepointer(s, freelist, old.freelist);
2119 		new.freelist = freelist;
2120 	} else
2121 		new.freelist = old.freelist;
2122 
2123 	new.frozen = 0;
2124 
2125 	if (!new.inuse && n->nr_partial >= s->min_partial)
2126 		m = M_FREE;
2127 	else if (new.freelist) {
2128 		m = M_PARTIAL;
2129 		if (!lock) {
2130 			lock = 1;
2131 			/*
2132 			 * Taking the spinlock removes the possibility
2133 			 * that acquire_slab() will see a slab page that
2134 			 * is frozen
2135 			 */
2136 			spin_lock(&n->list_lock);
2137 		}
2138 	} else {
2139 		m = M_FULL;
2140 		if (kmem_cache_debug(s) && !lock) {
2141 			lock = 1;
2142 			/*
2143 			 * This also ensures that the scanning of full
2144 			 * slabs from diagnostic functions will not see
2145 			 * any frozen slabs.
2146 			 */
2147 			spin_lock(&n->list_lock);
2148 		}
2149 	}
2150 
2151 	if (l != m) {
2152 		if (l == M_PARTIAL)
2153 			remove_partial(n, page);
2154 		else if (l == M_FULL)
2155 			remove_full(s, n, page);
2156 
2157 		if (m == M_PARTIAL)
2158 			add_partial(n, page, tail);
2159 		else if (m == M_FULL)
2160 			add_full(s, n, page);
2161 	}
2162 
2163 	l = m;
2164 	if (!__cmpxchg_double_slab(s, page,
2165 				old.freelist, old.counters,
2166 				new.freelist, new.counters,
2167 				"unfreezing slab"))
2168 		goto redo;
2169 
2170 	if (lock)
2171 		spin_unlock(&n->list_lock);
2172 
2173 	if (m == M_PARTIAL)
2174 		stat(s, tail);
2175 	else if (m == M_FULL)
2176 		stat(s, DEACTIVATE_FULL);
2177 	else if (m == M_FREE) {
2178 		stat(s, DEACTIVATE_EMPTY);
2179 		discard_slab(s, page);
2180 		stat(s, FREE_SLAB);
2181 	}
2182 
2183 	c->page = NULL;
2184 	c->freelist = NULL;
2185 }
2186 
2187 /*
2188  * Unfreeze all the cpu partial slabs.
2189  *
2190  * This function must be called with interrupts disabled
2191  * for the cpu using c (or some other guarantee must be there
2192  * to guarantee no concurrent accesses).
2193  */
2194 static void unfreeze_partials(struct kmem_cache *s,
2195 		struct kmem_cache_cpu *c)
2196 {
2197 #ifdef CONFIG_SLUB_CPU_PARTIAL
2198 	struct kmem_cache_node *n = NULL, *n2 = NULL;
2199 	struct page *page, *discard_page = NULL;
2200 
2201 	while ((page = c->partial)) {
2202 		struct page new;
2203 		struct page old;
2204 
2205 		c->partial = page->next;
2206 
2207 		n2 = get_node(s, page_to_nid(page));
2208 		if (n != n2) {
2209 			if (n)
2210 				spin_unlock(&n->list_lock);
2211 
2212 			n = n2;
2213 			spin_lock(&n->list_lock);
2214 		}
2215 
2216 		do {
2217 
2218 			old.freelist = page->freelist;
2219 			old.counters = page->counters;
2220 			VM_BUG_ON(!old.frozen);
2221 
2222 			new.counters = old.counters;
2223 			new.freelist = old.freelist;
2224 
2225 			new.frozen = 0;
2226 
2227 		} while (!__cmpxchg_double_slab(s, page,
2228 				old.freelist, old.counters,
2229 				new.freelist, new.counters,
2230 				"unfreezing slab"));
2231 
2232 		if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2233 			page->next = discard_page;
2234 			discard_page = page;
2235 		} else {
2236 			add_partial(n, page, DEACTIVATE_TO_TAIL);
2237 			stat(s, FREE_ADD_PARTIAL);
2238 		}
2239 	}
2240 
2241 	if (n)
2242 		spin_unlock(&n->list_lock);
2243 
2244 	while (discard_page) {
2245 		page = discard_page;
2246 		discard_page = discard_page->next;
2247 
2248 		stat(s, DEACTIVATE_EMPTY);
2249 		discard_slab(s, page);
2250 		stat(s, FREE_SLAB);
2251 	}
2252 #endif
2253 }
2254 
2255 /*
2256  * Put a page that was just frozen (in __slab_free|get_partial_node) into a
2257  * partial page slot if available.
2258  *
2259  * If we did not find a slot then simply move all the partials to the
2260  * per node partial list.
2261  */
2262 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
2263 {
2264 #ifdef CONFIG_SLUB_CPU_PARTIAL
2265 	struct page *oldpage;
2266 	int pages;
2267 	int pobjects;
2268 
2269 	preempt_disable();
2270 	do {
2271 		pages = 0;
2272 		pobjects = 0;
2273 		oldpage = this_cpu_read(s->cpu_slab->partial);
2274 
2275 		if (oldpage) {
2276 			pobjects = oldpage->pobjects;
2277 			pages = oldpage->pages;
2278 			if (drain && pobjects > s->cpu_partial) {
2279 				unsigned long flags;
2280 				/*
2281 				 * partial array is full. Move the existing
2282 				 * set to the per node partial list.
2283 				 */
2284 				local_irq_save(flags);
2285 				unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2286 				local_irq_restore(flags);
2287 				oldpage = NULL;
2288 				pobjects = 0;
2289 				pages = 0;
2290 				stat(s, CPU_PARTIAL_DRAIN);
2291 			}
2292 		}
2293 
2294 		pages++;
2295 		pobjects += page->objects - page->inuse;
2296 
2297 		page->pages = pages;
2298 		page->pobjects = pobjects;
2299 		page->next = oldpage;
2300 
2301 	} while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2302 								!= oldpage);
2303 	if (unlikely(!s->cpu_partial)) {
2304 		unsigned long flags;
2305 
2306 		local_irq_save(flags);
2307 		unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2308 		local_irq_restore(flags);
2309 	}
2310 	preempt_enable();
2311 #endif
2312 }
2313 
2314 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2315 {
2316 	stat(s, CPUSLAB_FLUSH);
2317 	deactivate_slab(s, c->page, c->freelist, c);
2318 
2319 	c->tid = next_tid(c->tid);
2320 }
2321 
2322 /*
2323  * Flush cpu slab.
2324  *
2325  * Called from IPI handler with interrupts disabled.
2326  */
2327 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2328 {
2329 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2330 
2331 	if (c->page)
2332 		flush_slab(s, c);
2333 
2334 	unfreeze_partials(s, c);
2335 }
2336 
2337 static void flush_cpu_slab(void *d)
2338 {
2339 	struct kmem_cache *s = d;
2340 
2341 	__flush_cpu_slab(s, smp_processor_id());
2342 }
2343 
2344 static bool has_cpu_slab(int cpu, void *info)
2345 {
2346 	struct kmem_cache *s = info;
2347 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2348 
2349 	return c->page || slub_percpu_partial(c);
2350 }
2351 
2352 static void flush_all(struct kmem_cache *s)
2353 {
2354 	on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
2355 }
2356 
2357 /*
2358  * Use the cpu notifier to insure that the cpu slabs are flushed when
2359  * necessary.
2360  */
2361 static int slub_cpu_dead(unsigned int cpu)
2362 {
2363 	struct kmem_cache *s;
2364 	unsigned long flags;
2365 
2366 	mutex_lock(&slab_mutex);
2367 	list_for_each_entry(s, &slab_caches, list) {
2368 		local_irq_save(flags);
2369 		__flush_cpu_slab(s, cpu);
2370 		local_irq_restore(flags);
2371 	}
2372 	mutex_unlock(&slab_mutex);
2373 	return 0;
2374 }
2375 
2376 /*
2377  * Check if the objects in a per cpu structure fit numa
2378  * locality expectations.
2379  */
2380 static inline int node_match(struct page *page, int node)
2381 {
2382 #ifdef CONFIG_NUMA
2383 	if (node != NUMA_NO_NODE && page_to_nid(page) != node)
2384 		return 0;
2385 #endif
2386 	return 1;
2387 }
2388 
2389 #ifdef CONFIG_SLUB_DEBUG
2390 static int count_free(struct page *page)
2391 {
2392 	return page->objects - page->inuse;
2393 }
2394 
2395 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2396 {
2397 	return atomic_long_read(&n->total_objects);
2398 }
2399 #endif /* CONFIG_SLUB_DEBUG */
2400 
2401 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2402 static unsigned long count_partial(struct kmem_cache_node *n,
2403 					int (*get_count)(struct page *))
2404 {
2405 	unsigned long flags;
2406 	unsigned long x = 0;
2407 	struct page *page;
2408 
2409 	spin_lock_irqsave(&n->list_lock, flags);
2410 	list_for_each_entry(page, &n->partial, lru)
2411 		x += get_count(page);
2412 	spin_unlock_irqrestore(&n->list_lock, flags);
2413 	return x;
2414 }
2415 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2416 
2417 static noinline void
2418 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2419 {
2420 #ifdef CONFIG_SLUB_DEBUG
2421 	static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2422 				      DEFAULT_RATELIMIT_BURST);
2423 	int node;
2424 	struct kmem_cache_node *n;
2425 
2426 	if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2427 		return;
2428 
2429 	pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2430 		nid, gfpflags, &gfpflags);
2431 	pr_warn("  cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2432 		s->name, s->object_size, s->size, oo_order(s->oo),
2433 		oo_order(s->min));
2434 
2435 	if (oo_order(s->min) > get_order(s->object_size))
2436 		pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2437 			s->name);
2438 
2439 	for_each_kmem_cache_node(s, node, n) {
2440 		unsigned long nr_slabs;
2441 		unsigned long nr_objs;
2442 		unsigned long nr_free;
2443 
2444 		nr_free  = count_partial(n, count_free);
2445 		nr_slabs = node_nr_slabs(n);
2446 		nr_objs  = node_nr_objs(n);
2447 
2448 		pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2449 			node, nr_slabs, nr_objs, nr_free);
2450 	}
2451 #endif
2452 }
2453 
2454 static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2455 			int node, struct kmem_cache_cpu **pc)
2456 {
2457 	void *freelist;
2458 	struct kmem_cache_cpu *c = *pc;
2459 	struct page *page;
2460 
2461 	WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2462 
2463 	freelist = get_partial(s, flags, node, c);
2464 
2465 	if (freelist)
2466 		return freelist;
2467 
2468 	page = new_slab(s, flags, node);
2469 	if (page) {
2470 		c = raw_cpu_ptr(s->cpu_slab);
2471 		if (c->page)
2472 			flush_slab(s, c);
2473 
2474 		/*
2475 		 * No other reference to the page yet so we can
2476 		 * muck around with it freely without cmpxchg
2477 		 */
2478 		freelist = page->freelist;
2479 		page->freelist = NULL;
2480 
2481 		stat(s, ALLOC_SLAB);
2482 		c->page = page;
2483 		*pc = c;
2484 	}
2485 
2486 	return freelist;
2487 }
2488 
2489 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2490 {
2491 	if (unlikely(PageSlabPfmemalloc(page)))
2492 		return gfp_pfmemalloc_allowed(gfpflags);
2493 
2494 	return true;
2495 }
2496 
2497 /*
2498  * Check the page->freelist of a page and either transfer the freelist to the
2499  * per cpu freelist or deactivate the page.
2500  *
2501  * The page is still frozen if the return value is not NULL.
2502  *
2503  * If this function returns NULL then the page has been unfrozen.
2504  *
2505  * This function must be called with interrupt disabled.
2506  */
2507 static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2508 {
2509 	struct page new;
2510 	unsigned long counters;
2511 	void *freelist;
2512 
2513 	do {
2514 		freelist = page->freelist;
2515 		counters = page->counters;
2516 
2517 		new.counters = counters;
2518 		VM_BUG_ON(!new.frozen);
2519 
2520 		new.inuse = page->objects;
2521 		new.frozen = freelist != NULL;
2522 
2523 	} while (!__cmpxchg_double_slab(s, page,
2524 		freelist, counters,
2525 		NULL, new.counters,
2526 		"get_freelist"));
2527 
2528 	return freelist;
2529 }
2530 
2531 /*
2532  * Slow path. The lockless freelist is empty or we need to perform
2533  * debugging duties.
2534  *
2535  * Processing is still very fast if new objects have been freed to the
2536  * regular freelist. In that case we simply take over the regular freelist
2537  * as the lockless freelist and zap the regular freelist.
2538  *
2539  * If that is not working then we fall back to the partial lists. We take the
2540  * first element of the freelist as the object to allocate now and move the
2541  * rest of the freelist to the lockless freelist.
2542  *
2543  * And if we were unable to get a new slab from the partial slab lists then
2544  * we need to allocate a new slab. This is the slowest path since it involves
2545  * a call to the page allocator and the setup of a new slab.
2546  *
2547  * Version of __slab_alloc to use when we know that interrupts are
2548  * already disabled (which is the case for bulk allocation).
2549  */
2550 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2551 			  unsigned long addr, struct kmem_cache_cpu *c)
2552 {
2553 	void *freelist;
2554 	struct page *page;
2555 
2556 	page = c->page;
2557 	if (!page)
2558 		goto new_slab;
2559 redo:
2560 
2561 	if (unlikely(!node_match(page, node))) {
2562 		int searchnode = node;
2563 
2564 		if (node != NUMA_NO_NODE && !node_present_pages(node))
2565 			searchnode = node_to_mem_node(node);
2566 
2567 		if (unlikely(!node_match(page, searchnode))) {
2568 			stat(s, ALLOC_NODE_MISMATCH);
2569 			deactivate_slab(s, page, c->freelist, c);
2570 			goto new_slab;
2571 		}
2572 	}
2573 
2574 	/*
2575 	 * By rights, we should be searching for a slab page that was
2576 	 * PFMEMALLOC but right now, we are losing the pfmemalloc
2577 	 * information when the page leaves the per-cpu allocator
2578 	 */
2579 	if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2580 		deactivate_slab(s, page, c->freelist, c);
2581 		goto new_slab;
2582 	}
2583 
2584 	/* must check again c->freelist in case of cpu migration or IRQ */
2585 	freelist = c->freelist;
2586 	if (freelist)
2587 		goto load_freelist;
2588 
2589 	freelist = get_freelist(s, page);
2590 
2591 	if (!freelist) {
2592 		c->page = NULL;
2593 		stat(s, DEACTIVATE_BYPASS);
2594 		goto new_slab;
2595 	}
2596 
2597 	stat(s, ALLOC_REFILL);
2598 
2599 load_freelist:
2600 	/*
2601 	 * freelist is pointing to the list of objects to be used.
2602 	 * page is pointing to the page from which the objects are obtained.
2603 	 * That page must be frozen for per cpu allocations to work.
2604 	 */
2605 	VM_BUG_ON(!c->page->frozen);
2606 	c->freelist = get_freepointer(s, freelist);
2607 	c->tid = next_tid(c->tid);
2608 	return freelist;
2609 
2610 new_slab:
2611 
2612 	if (slub_percpu_partial(c)) {
2613 		page = c->page = slub_percpu_partial(c);
2614 		slub_set_percpu_partial(c, page);
2615 		stat(s, CPU_PARTIAL_ALLOC);
2616 		goto redo;
2617 	}
2618 
2619 	freelist = new_slab_objects(s, gfpflags, node, &c);
2620 
2621 	if (unlikely(!freelist)) {
2622 		slab_out_of_memory(s, gfpflags, node);
2623 		return NULL;
2624 	}
2625 
2626 	page = c->page;
2627 	if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
2628 		goto load_freelist;
2629 
2630 	/* Only entered in the debug case */
2631 	if (kmem_cache_debug(s) &&
2632 			!alloc_debug_processing(s, page, freelist, addr))
2633 		goto new_slab;	/* Slab failed checks. Next slab needed */
2634 
2635 	deactivate_slab(s, page, get_freepointer(s, freelist), c);
2636 	return freelist;
2637 }
2638 
2639 /*
2640  * Another one that disabled interrupt and compensates for possible
2641  * cpu changes by refetching the per cpu area pointer.
2642  */
2643 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2644 			  unsigned long addr, struct kmem_cache_cpu *c)
2645 {
2646 	void *p;
2647 	unsigned long flags;
2648 
2649 	local_irq_save(flags);
2650 #ifdef CONFIG_PREEMPT
2651 	/*
2652 	 * We may have been preempted and rescheduled on a different
2653 	 * cpu before disabling interrupts. Need to reload cpu area
2654 	 * pointer.
2655 	 */
2656 	c = this_cpu_ptr(s->cpu_slab);
2657 #endif
2658 
2659 	p = ___slab_alloc(s, gfpflags, node, addr, c);
2660 	local_irq_restore(flags);
2661 	return p;
2662 }
2663 
2664 /*
2665  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2666  * have the fastpath folded into their functions. So no function call
2667  * overhead for requests that can be satisfied on the fastpath.
2668  *
2669  * The fastpath works by first checking if the lockless freelist can be used.
2670  * If not then __slab_alloc is called for slow processing.
2671  *
2672  * Otherwise we can simply pick the next object from the lockless free list.
2673  */
2674 static __always_inline void *slab_alloc_node(struct kmem_cache *s,
2675 		gfp_t gfpflags, int node, unsigned long addr)
2676 {
2677 	void *object;
2678 	struct kmem_cache_cpu *c;
2679 	struct page *page;
2680 	unsigned long tid;
2681 
2682 	s = slab_pre_alloc_hook(s, gfpflags);
2683 	if (!s)
2684 		return NULL;
2685 redo:
2686 	/*
2687 	 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2688 	 * enabled. We may switch back and forth between cpus while
2689 	 * reading from one cpu area. That does not matter as long
2690 	 * as we end up on the original cpu again when doing the cmpxchg.
2691 	 *
2692 	 * We should guarantee that tid and kmem_cache are retrieved on
2693 	 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2694 	 * to check if it is matched or not.
2695 	 */
2696 	do {
2697 		tid = this_cpu_read(s->cpu_slab->tid);
2698 		c = raw_cpu_ptr(s->cpu_slab);
2699 	} while (IS_ENABLED(CONFIG_PREEMPT) &&
2700 		 unlikely(tid != READ_ONCE(c->tid)));
2701 
2702 	/*
2703 	 * Irqless object alloc/free algorithm used here depends on sequence
2704 	 * of fetching cpu_slab's data. tid should be fetched before anything
2705 	 * on c to guarantee that object and page associated with previous tid
2706 	 * won't be used with current tid. If we fetch tid first, object and
2707 	 * page could be one associated with next tid and our alloc/free
2708 	 * request will be failed. In this case, we will retry. So, no problem.
2709 	 */
2710 	barrier();
2711 
2712 	/*
2713 	 * The transaction ids are globally unique per cpu and per operation on
2714 	 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2715 	 * occurs on the right processor and that there was no operation on the
2716 	 * linked list in between.
2717 	 */
2718 
2719 	object = c->freelist;
2720 	page = c->page;
2721 	if (unlikely(!object || !node_match(page, node))) {
2722 		object = __slab_alloc(s, gfpflags, node, addr, c);
2723 		stat(s, ALLOC_SLOWPATH);
2724 	} else {
2725 		void *next_object = get_freepointer_safe(s, object);
2726 
2727 		/*
2728 		 * The cmpxchg will only match if there was no additional
2729 		 * operation and if we are on the right processor.
2730 		 *
2731 		 * The cmpxchg does the following atomically (without lock
2732 		 * semantics!)
2733 		 * 1. Relocate first pointer to the current per cpu area.
2734 		 * 2. Verify that tid and freelist have not been changed
2735 		 * 3. If they were not changed replace tid and freelist
2736 		 *
2737 		 * Since this is without lock semantics the protection is only
2738 		 * against code executing on this cpu *not* from access by
2739 		 * other cpus.
2740 		 */
2741 		if (unlikely(!this_cpu_cmpxchg_double(
2742 				s->cpu_slab->freelist, s->cpu_slab->tid,
2743 				object, tid,
2744 				next_object, next_tid(tid)))) {
2745 
2746 			note_cmpxchg_failure("slab_alloc", s, tid);
2747 			goto redo;
2748 		}
2749 		prefetch_freepointer(s, next_object);
2750 		stat(s, ALLOC_FASTPATH);
2751 	}
2752 
2753 	if (unlikely(gfpflags & __GFP_ZERO) && object)
2754 		memset(object, 0, s->object_size);
2755 
2756 	slab_post_alloc_hook(s, gfpflags, 1, &object);
2757 
2758 	return object;
2759 }
2760 
2761 static __always_inline void *slab_alloc(struct kmem_cache *s,
2762 		gfp_t gfpflags, unsigned long addr)
2763 {
2764 	return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2765 }
2766 
2767 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2768 {
2769 	void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2770 
2771 	trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2772 				s->size, gfpflags);
2773 
2774 	return ret;
2775 }
2776 EXPORT_SYMBOL(kmem_cache_alloc);
2777 
2778 #ifdef CONFIG_TRACING
2779 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2780 {
2781 	void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2782 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2783 	ret = kasan_kmalloc(s, ret, size, gfpflags);
2784 	return ret;
2785 }
2786 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2787 #endif
2788 
2789 #ifdef CONFIG_NUMA
2790 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2791 {
2792 	void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2793 
2794 	trace_kmem_cache_alloc_node(_RET_IP_, ret,
2795 				    s->object_size, s->size, gfpflags, node);
2796 
2797 	return ret;
2798 }
2799 EXPORT_SYMBOL(kmem_cache_alloc_node);
2800 
2801 #ifdef CONFIG_TRACING
2802 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2803 				    gfp_t gfpflags,
2804 				    int node, size_t size)
2805 {
2806 	void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2807 
2808 	trace_kmalloc_node(_RET_IP_, ret,
2809 			   size, s->size, gfpflags, node);
2810 
2811 	ret = kasan_kmalloc(s, ret, size, gfpflags);
2812 	return ret;
2813 }
2814 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2815 #endif
2816 #endif
2817 
2818 /*
2819  * Slow path handling. This may still be called frequently since objects
2820  * have a longer lifetime than the cpu slabs in most processing loads.
2821  *
2822  * So we still attempt to reduce cache line usage. Just take the slab
2823  * lock and free the item. If there is no additional partial page
2824  * handling required then we can return immediately.
2825  */
2826 static void __slab_free(struct kmem_cache *s, struct page *page,
2827 			void *head, void *tail, int cnt,
2828 			unsigned long addr)
2829 
2830 {
2831 	void *prior;
2832 	int was_frozen;
2833 	struct page new;
2834 	unsigned long counters;
2835 	struct kmem_cache_node *n = NULL;
2836 	unsigned long uninitialized_var(flags);
2837 
2838 	stat(s, FREE_SLOWPATH);
2839 
2840 	if (kmem_cache_debug(s) &&
2841 	    !free_debug_processing(s, page, head, tail, cnt, addr))
2842 		return;
2843 
2844 	do {
2845 		if (unlikely(n)) {
2846 			spin_unlock_irqrestore(&n->list_lock, flags);
2847 			n = NULL;
2848 		}
2849 		prior = page->freelist;
2850 		counters = page->counters;
2851 		set_freepointer(s, tail, prior);
2852 		new.counters = counters;
2853 		was_frozen = new.frozen;
2854 		new.inuse -= cnt;
2855 		if ((!new.inuse || !prior) && !was_frozen) {
2856 
2857 			if (kmem_cache_has_cpu_partial(s) && !prior) {
2858 
2859 				/*
2860 				 * Slab was on no list before and will be
2861 				 * partially empty
2862 				 * We can defer the list move and instead
2863 				 * freeze it.
2864 				 */
2865 				new.frozen = 1;
2866 
2867 			} else { /* Needs to be taken off a list */
2868 
2869 				n = get_node(s, page_to_nid(page));
2870 				/*
2871 				 * Speculatively acquire the list_lock.
2872 				 * If the cmpxchg does not succeed then we may
2873 				 * drop the list_lock without any processing.
2874 				 *
2875 				 * Otherwise the list_lock will synchronize with
2876 				 * other processors updating the list of slabs.
2877 				 */
2878 				spin_lock_irqsave(&n->list_lock, flags);
2879 
2880 			}
2881 		}
2882 
2883 	} while (!cmpxchg_double_slab(s, page,
2884 		prior, counters,
2885 		head, new.counters,
2886 		"__slab_free"));
2887 
2888 	if (likely(!n)) {
2889 
2890 		/*
2891 		 * If we just froze the page then put it onto the
2892 		 * per cpu partial list.
2893 		 */
2894 		if (new.frozen && !was_frozen) {
2895 			put_cpu_partial(s, page, 1);
2896 			stat(s, CPU_PARTIAL_FREE);
2897 		}
2898 		/*
2899 		 * The list lock was not taken therefore no list
2900 		 * activity can be necessary.
2901 		 */
2902 		if (was_frozen)
2903 			stat(s, FREE_FROZEN);
2904 		return;
2905 	}
2906 
2907 	if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
2908 		goto slab_empty;
2909 
2910 	/*
2911 	 * Objects left in the slab. If it was not on the partial list before
2912 	 * then add it.
2913 	 */
2914 	if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2915 		if (kmem_cache_debug(s))
2916 			remove_full(s, n, page);
2917 		add_partial(n, page, DEACTIVATE_TO_TAIL);
2918 		stat(s, FREE_ADD_PARTIAL);
2919 	}
2920 	spin_unlock_irqrestore(&n->list_lock, flags);
2921 	return;
2922 
2923 slab_empty:
2924 	if (prior) {
2925 		/*
2926 		 * Slab on the partial list.
2927 		 */
2928 		remove_partial(n, page);
2929 		stat(s, FREE_REMOVE_PARTIAL);
2930 	} else {
2931 		/* Slab must be on the full list */
2932 		remove_full(s, n, page);
2933 	}
2934 
2935 	spin_unlock_irqrestore(&n->list_lock, flags);
2936 	stat(s, FREE_SLAB);
2937 	discard_slab(s, page);
2938 }
2939 
2940 /*
2941  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2942  * can perform fastpath freeing without additional function calls.
2943  *
2944  * The fastpath is only possible if we are freeing to the current cpu slab
2945  * of this processor. This typically the case if we have just allocated
2946  * the item before.
2947  *
2948  * If fastpath is not possible then fall back to __slab_free where we deal
2949  * with all sorts of special processing.
2950  *
2951  * Bulk free of a freelist with several objects (all pointing to the
2952  * same page) possible by specifying head and tail ptr, plus objects
2953  * count (cnt). Bulk free indicated by tail pointer being set.
2954  */
2955 static __always_inline void do_slab_free(struct kmem_cache *s,
2956 				struct page *page, void *head, void *tail,
2957 				int cnt, unsigned long addr)
2958 {
2959 	void *tail_obj = tail ? : head;
2960 	struct kmem_cache_cpu *c;
2961 	unsigned long tid;
2962 redo:
2963 	/*
2964 	 * Determine the currently cpus per cpu slab.
2965 	 * The cpu may change afterward. However that does not matter since
2966 	 * data is retrieved via this pointer. If we are on the same cpu
2967 	 * during the cmpxchg then the free will succeed.
2968 	 */
2969 	do {
2970 		tid = this_cpu_read(s->cpu_slab->tid);
2971 		c = raw_cpu_ptr(s->cpu_slab);
2972 	} while (IS_ENABLED(CONFIG_PREEMPT) &&
2973 		 unlikely(tid != READ_ONCE(c->tid)));
2974 
2975 	/* Same with comment on barrier() in slab_alloc_node() */
2976 	barrier();
2977 
2978 	if (likely(page == c->page)) {
2979 		set_freepointer(s, tail_obj, c->freelist);
2980 
2981 		if (unlikely(!this_cpu_cmpxchg_double(
2982 				s->cpu_slab->freelist, s->cpu_slab->tid,
2983 				c->freelist, tid,
2984 				head, next_tid(tid)))) {
2985 
2986 			note_cmpxchg_failure("slab_free", s, tid);
2987 			goto redo;
2988 		}
2989 		stat(s, FREE_FASTPATH);
2990 	} else
2991 		__slab_free(s, page, head, tail_obj, cnt, addr);
2992 
2993 }
2994 
2995 static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2996 				      void *head, void *tail, int cnt,
2997 				      unsigned long addr)
2998 {
2999 	/*
3000 	 * With KASAN enabled slab_free_freelist_hook modifies the freelist
3001 	 * to remove objects, whose reuse must be delayed.
3002 	 */
3003 	if (slab_free_freelist_hook(s, &head, &tail))
3004 		do_slab_free(s, page, head, tail, cnt, addr);
3005 }
3006 
3007 #ifdef CONFIG_KASAN_GENERIC
3008 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3009 {
3010 	do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
3011 }
3012 #endif
3013 
3014 void kmem_cache_free(struct kmem_cache *s, void *x)
3015 {
3016 	s = cache_from_obj(s, x);
3017 	if (!s)
3018 		return;
3019 	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
3020 	trace_kmem_cache_free(_RET_IP_, x);
3021 }
3022 EXPORT_SYMBOL(kmem_cache_free);
3023 
3024 struct detached_freelist {
3025 	struct page *page;
3026 	void *tail;
3027 	void *freelist;
3028 	int cnt;
3029 	struct kmem_cache *s;
3030 };
3031 
3032 /*
3033  * This function progressively scans the array with free objects (with
3034  * a limited look ahead) and extract objects belonging to the same
3035  * page.  It builds a detached freelist directly within the given
3036  * page/objects.  This can happen without any need for
3037  * synchronization, because the objects are owned by running process.
3038  * The freelist is build up as a single linked list in the objects.
3039  * The idea is, that this detached freelist can then be bulk
3040  * transferred to the real freelist(s), but only requiring a single
3041  * synchronization primitive.  Look ahead in the array is limited due
3042  * to performance reasons.
3043  */
3044 static inline
3045 int build_detached_freelist(struct kmem_cache *s, size_t size,
3046 			    void **p, struct detached_freelist *df)
3047 {
3048 	size_t first_skipped_index = 0;
3049 	int lookahead = 3;
3050 	void *object;
3051 	struct page *page;
3052 
3053 	/* Always re-init detached_freelist */
3054 	df->page = NULL;
3055 
3056 	do {
3057 		object = p[--size];
3058 		/* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
3059 	} while (!object && size);
3060 
3061 	if (!object)
3062 		return 0;
3063 
3064 	page = virt_to_head_page(object);
3065 	if (!s) {
3066 		/* Handle kalloc'ed objects */
3067 		if (unlikely(!PageSlab(page))) {
3068 			BUG_ON(!PageCompound(page));
3069 			kfree_hook(object);
3070 			__free_pages(page, compound_order(page));
3071 			p[size] = NULL; /* mark object processed */
3072 			return size;
3073 		}
3074 		/* Derive kmem_cache from object */
3075 		df->s = page->slab_cache;
3076 	} else {
3077 		df->s = cache_from_obj(s, object); /* Support for memcg */
3078 	}
3079 
3080 	/* Start new detached freelist */
3081 	df->page = page;
3082 	set_freepointer(df->s, object, NULL);
3083 	df->tail = object;
3084 	df->freelist = object;
3085 	p[size] = NULL; /* mark object processed */
3086 	df->cnt = 1;
3087 
3088 	while (size) {
3089 		object = p[--size];
3090 		if (!object)
3091 			continue; /* Skip processed objects */
3092 
3093 		/* df->page is always set at this point */
3094 		if (df->page == virt_to_head_page(object)) {
3095 			/* Opportunity build freelist */
3096 			set_freepointer(df->s, object, df->freelist);
3097 			df->freelist = object;
3098 			df->cnt++;
3099 			p[size] = NULL; /* mark object processed */
3100 
3101 			continue;
3102 		}
3103 
3104 		/* Limit look ahead search */
3105 		if (!--lookahead)
3106 			break;
3107 
3108 		if (!first_skipped_index)
3109 			first_skipped_index = size + 1;
3110 	}
3111 
3112 	return first_skipped_index;
3113 }
3114 
3115 /* Note that interrupts must be enabled when calling this function. */
3116 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3117 {
3118 	if (WARN_ON(!size))
3119 		return;
3120 
3121 	do {
3122 		struct detached_freelist df;
3123 
3124 		size = build_detached_freelist(s, size, p, &df);
3125 		if (!df.page)
3126 			continue;
3127 
3128 		slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
3129 	} while (likely(size));
3130 }
3131 EXPORT_SYMBOL(kmem_cache_free_bulk);
3132 
3133 /* Note that interrupts must be enabled when calling this function. */
3134 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3135 			  void **p)
3136 {
3137 	struct kmem_cache_cpu *c;
3138 	int i;
3139 
3140 	/* memcg and kmem_cache debug support */
3141 	s = slab_pre_alloc_hook(s, flags);
3142 	if (unlikely(!s))
3143 		return false;
3144 	/*
3145 	 * Drain objects in the per cpu slab, while disabling local
3146 	 * IRQs, which protects against PREEMPT and interrupts
3147 	 * handlers invoking normal fastpath.
3148 	 */
3149 	local_irq_disable();
3150 	c = this_cpu_ptr(s->cpu_slab);
3151 
3152 	for (i = 0; i < size; i++) {
3153 		void *object = c->freelist;
3154 
3155 		if (unlikely(!object)) {
3156 			/*
3157 			 * Invoking slow path likely have side-effect
3158 			 * of re-populating per CPU c->freelist
3159 			 */
3160 			p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3161 					    _RET_IP_, c);
3162 			if (unlikely(!p[i]))
3163 				goto error;
3164 
3165 			c = this_cpu_ptr(s->cpu_slab);
3166 			continue; /* goto for-loop */
3167 		}
3168 		c->freelist = get_freepointer(s, object);
3169 		p[i] = object;
3170 	}
3171 	c->tid = next_tid(c->tid);
3172 	local_irq_enable();
3173 
3174 	/* Clear memory outside IRQ disabled fastpath loop */
3175 	if (unlikely(flags & __GFP_ZERO)) {
3176 		int j;
3177 
3178 		for (j = 0; j < i; j++)
3179 			memset(p[j], 0, s->object_size);
3180 	}
3181 
3182 	/* memcg and kmem_cache debug support */
3183 	slab_post_alloc_hook(s, flags, size, p);
3184 	return i;
3185 error:
3186 	local_irq_enable();
3187 	slab_post_alloc_hook(s, flags, i, p);
3188 	__kmem_cache_free_bulk(s, i, p);
3189 	return 0;
3190 }
3191 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3192 
3193 
3194 /*
3195  * Object placement in a slab is made very easy because we always start at
3196  * offset 0. If we tune the size of the object to the alignment then we can
3197  * get the required alignment by putting one properly sized object after
3198  * another.
3199  *
3200  * Notice that the allocation order determines the sizes of the per cpu
3201  * caches. Each processor has always one slab available for allocations.
3202  * Increasing the allocation order reduces the number of times that slabs
3203  * must be moved on and off the partial lists and is therefore a factor in
3204  * locking overhead.
3205  */
3206 
3207 /*
3208  * Mininum / Maximum order of slab pages. This influences locking overhead
3209  * and slab fragmentation. A higher order reduces the number of partial slabs
3210  * and increases the number of allocations possible without having to
3211  * take the list_lock.
3212  */
3213 static unsigned int slub_min_order;
3214 static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3215 static unsigned int slub_min_objects;
3216 
3217 /*
3218  * Calculate the order of allocation given an slab object size.
3219  *
3220  * The order of allocation has significant impact on performance and other
3221  * system components. Generally order 0 allocations should be preferred since
3222  * order 0 does not cause fragmentation in the page allocator. Larger objects
3223  * be problematic to put into order 0 slabs because there may be too much
3224  * unused space left. We go to a higher order if more than 1/16th of the slab
3225  * would be wasted.
3226  *
3227  * In order to reach satisfactory performance we must ensure that a minimum
3228  * number of objects is in one slab. Otherwise we may generate too much
3229  * activity on the partial lists which requires taking the list_lock. This is
3230  * less a concern for large slabs though which are rarely used.
3231  *
3232  * slub_max_order specifies the order where we begin to stop considering the
3233  * number of objects in a slab as critical. If we reach slub_max_order then
3234  * we try to keep the page order as low as possible. So we accept more waste
3235  * of space in favor of a small page order.
3236  *
3237  * Higher order allocations also allow the placement of more objects in a
3238  * slab and thereby reduce object handling overhead. If the user has
3239  * requested a higher mininum order then we start with that one instead of
3240  * the smallest order which will fit the object.
3241  */
3242 static inline unsigned int slab_order(unsigned int size,
3243 		unsigned int min_objects, unsigned int max_order,
3244 		unsigned int fract_leftover)
3245 {
3246 	unsigned int min_order = slub_min_order;
3247 	unsigned int order;
3248 
3249 	if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
3250 		return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3251 
3252 	for (order = max(min_order, (unsigned int)get_order(min_objects * size));
3253 			order <= max_order; order++) {
3254 
3255 		unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3256 		unsigned int rem;
3257 
3258 		rem = slab_size % size;
3259 
3260 		if (rem <= slab_size / fract_leftover)
3261 			break;
3262 	}
3263 
3264 	return order;
3265 }
3266 
3267 static inline int calculate_order(unsigned int size)
3268 {
3269 	unsigned int order;
3270 	unsigned int min_objects;
3271 	unsigned int max_objects;
3272 
3273 	/*
3274 	 * Attempt to find best configuration for a slab. This
3275 	 * works by first attempting to generate a layout with
3276 	 * the best configuration and backing off gradually.
3277 	 *
3278 	 * First we increase the acceptable waste in a slab. Then
3279 	 * we reduce the minimum objects required in a slab.
3280 	 */
3281 	min_objects = slub_min_objects;
3282 	if (!min_objects)
3283 		min_objects = 4 * (fls(nr_cpu_ids) + 1);
3284 	max_objects = order_objects(slub_max_order, size);
3285 	min_objects = min(min_objects, max_objects);
3286 
3287 	while (min_objects > 1) {
3288 		unsigned int fraction;
3289 
3290 		fraction = 16;
3291 		while (fraction >= 4) {
3292 			order = slab_order(size, min_objects,
3293 					slub_max_order, fraction);
3294 			if (order <= slub_max_order)
3295 				return order;
3296 			fraction /= 2;
3297 		}
3298 		min_objects--;
3299 	}
3300 
3301 	/*
3302 	 * We were unable to place multiple objects in a slab. Now
3303 	 * lets see if we can place a single object there.
3304 	 */
3305 	order = slab_order(size, 1, slub_max_order, 1);
3306 	if (order <= slub_max_order)
3307 		return order;
3308 
3309 	/*
3310 	 * Doh this slab cannot be placed using slub_max_order.
3311 	 */
3312 	order = slab_order(size, 1, MAX_ORDER, 1);
3313 	if (order < MAX_ORDER)
3314 		return order;
3315 	return -ENOSYS;
3316 }
3317 
3318 static void
3319 init_kmem_cache_node(struct kmem_cache_node *n)
3320 {
3321 	n->nr_partial = 0;
3322 	spin_lock_init(&n->list_lock);
3323 	INIT_LIST_HEAD(&n->partial);
3324 #ifdef CONFIG_SLUB_DEBUG
3325 	atomic_long_set(&n->nr_slabs, 0);
3326 	atomic_long_set(&n->total_objects, 0);
3327 	INIT_LIST_HEAD(&n->full);
3328 #endif
3329 }
3330 
3331 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3332 {
3333 	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3334 			KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3335 
3336 	/*
3337 	 * Must align to double word boundary for the double cmpxchg
3338 	 * instructions to work; see __pcpu_double_call_return_bool().
3339 	 */
3340 	s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3341 				     2 * sizeof(void *));
3342 
3343 	if (!s->cpu_slab)
3344 		return 0;
3345 
3346 	init_kmem_cache_cpus(s);
3347 
3348 	return 1;
3349 }
3350 
3351 static struct kmem_cache *kmem_cache_node;
3352 
3353 /*
3354  * No kmalloc_node yet so do it by hand. We know that this is the first
3355  * slab on the node for this slabcache. There are no concurrent accesses
3356  * possible.
3357  *
3358  * Note that this function only works on the kmem_cache_node
3359  * when allocating for the kmem_cache_node. This is used for bootstrapping
3360  * memory on a fresh node that has no slab structures yet.
3361  */
3362 static void early_kmem_cache_node_alloc(int node)
3363 {
3364 	struct page *page;
3365 	struct kmem_cache_node *n;
3366 
3367 	BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3368 
3369 	page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3370 
3371 	BUG_ON(!page);
3372 	if (page_to_nid(page) != node) {
3373 		pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3374 		pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3375 	}
3376 
3377 	n = page->freelist;
3378 	BUG_ON(!n);
3379 #ifdef CONFIG_SLUB_DEBUG
3380 	init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3381 	init_tracking(kmem_cache_node, n);
3382 #endif
3383 	n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3384 		      GFP_KERNEL);
3385 	page->freelist = get_freepointer(kmem_cache_node, n);
3386 	page->inuse = 1;
3387 	page->frozen = 0;
3388 	kmem_cache_node->node[node] = n;
3389 	init_kmem_cache_node(n);
3390 	inc_slabs_node(kmem_cache_node, node, page->objects);
3391 
3392 	/*
3393 	 * No locks need to be taken here as it has just been
3394 	 * initialized and there is no concurrent access.
3395 	 */
3396 	__add_partial(n, page, DEACTIVATE_TO_HEAD);
3397 }
3398 
3399 static void free_kmem_cache_nodes(struct kmem_cache *s)
3400 {
3401 	int node;
3402 	struct kmem_cache_node *n;
3403 
3404 	for_each_kmem_cache_node(s, node, n) {
3405 		s->node[node] = NULL;
3406 		kmem_cache_free(kmem_cache_node, n);
3407 	}
3408 }
3409 
3410 void __kmem_cache_release(struct kmem_cache *s)
3411 {
3412 	cache_random_seq_destroy(s);
3413 	free_percpu(s->cpu_slab);
3414 	free_kmem_cache_nodes(s);
3415 }
3416 
3417 static int init_kmem_cache_nodes(struct kmem_cache *s)
3418 {
3419 	int node;
3420 
3421 	for_each_node_state(node, N_NORMAL_MEMORY) {
3422 		struct kmem_cache_node *n;
3423 
3424 		if (slab_state == DOWN) {
3425 			early_kmem_cache_node_alloc(node);
3426 			continue;
3427 		}
3428 		n = kmem_cache_alloc_node(kmem_cache_node,
3429 						GFP_KERNEL, node);
3430 
3431 		if (!n) {
3432 			free_kmem_cache_nodes(s);
3433 			return 0;
3434 		}
3435 
3436 		init_kmem_cache_node(n);
3437 		s->node[node] = n;
3438 	}
3439 	return 1;
3440 }
3441 
3442 static void set_min_partial(struct kmem_cache *s, unsigned long min)
3443 {
3444 	if (min < MIN_PARTIAL)
3445 		min = MIN_PARTIAL;
3446 	else if (min > MAX_PARTIAL)
3447 		min = MAX_PARTIAL;
3448 	s->min_partial = min;
3449 }
3450 
3451 static void set_cpu_partial(struct kmem_cache *s)
3452 {
3453 #ifdef CONFIG_SLUB_CPU_PARTIAL
3454 	/*
3455 	 * cpu_partial determined the maximum number of objects kept in the
3456 	 * per cpu partial lists of a processor.
3457 	 *
3458 	 * Per cpu partial lists mainly contain slabs that just have one
3459 	 * object freed. If they are used for allocation then they can be
3460 	 * filled up again with minimal effort. The slab will never hit the
3461 	 * per node partial lists and therefore no locking will be required.
3462 	 *
3463 	 * This setting also determines
3464 	 *
3465 	 * A) The number of objects from per cpu partial slabs dumped to the
3466 	 *    per node list when we reach the limit.
3467 	 * B) The number of objects in cpu partial slabs to extract from the
3468 	 *    per node list when we run out of per cpu objects. We only fetch
3469 	 *    50% to keep some capacity around for frees.
3470 	 */
3471 	if (!kmem_cache_has_cpu_partial(s))
3472 		s->cpu_partial = 0;
3473 	else if (s->size >= PAGE_SIZE)
3474 		s->cpu_partial = 2;
3475 	else if (s->size >= 1024)
3476 		s->cpu_partial = 6;
3477 	else if (s->size >= 256)
3478 		s->cpu_partial = 13;
3479 	else
3480 		s->cpu_partial = 30;
3481 #endif
3482 }
3483 
3484 /*
3485  * calculate_sizes() determines the order and the distribution of data within
3486  * a slab object.
3487  */
3488 static int calculate_sizes(struct kmem_cache *s, int forced_order)
3489 {
3490 	slab_flags_t flags = s->flags;
3491 	unsigned int size = s->object_size;
3492 	unsigned int order;
3493 
3494 	/*
3495 	 * Round up object size to the next word boundary. We can only
3496 	 * place the free pointer at word boundaries and this determines
3497 	 * the possible location of the free pointer.
3498 	 */
3499 	size = ALIGN(size, sizeof(void *));
3500 
3501 #ifdef CONFIG_SLUB_DEBUG
3502 	/*
3503 	 * Determine if we can poison the object itself. If the user of
3504 	 * the slab may touch the object after free or before allocation
3505 	 * then we should never poison the object itself.
3506 	 */
3507 	if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
3508 			!s->ctor)
3509 		s->flags |= __OBJECT_POISON;
3510 	else
3511 		s->flags &= ~__OBJECT_POISON;
3512 
3513 
3514 	/*
3515 	 * If we are Redzoning then check if there is some space between the
3516 	 * end of the object and the free pointer. If not then add an
3517 	 * additional word to have some bytes to store Redzone information.
3518 	 */
3519 	if ((flags & SLAB_RED_ZONE) && size == s->object_size)
3520 		size += sizeof(void *);
3521 #endif
3522 
3523 	/*
3524 	 * With that we have determined the number of bytes in actual use
3525 	 * by the object. This is the potential offset to the free pointer.
3526 	 */
3527 	s->inuse = size;
3528 
3529 	if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
3530 		s->ctor)) {
3531 		/*
3532 		 * Relocate free pointer after the object if it is not
3533 		 * permitted to overwrite the first word of the object on
3534 		 * kmem_cache_free.
3535 		 *
3536 		 * This is the case if we do RCU, have a constructor or
3537 		 * destructor or are poisoning the objects.
3538 		 */
3539 		s->offset = size;
3540 		size += sizeof(void *);
3541 	}
3542 
3543 #ifdef CONFIG_SLUB_DEBUG
3544 	if (flags & SLAB_STORE_USER)
3545 		/*
3546 		 * Need to store information about allocs and frees after
3547 		 * the object.
3548 		 */
3549 		size += 2 * sizeof(struct track);
3550 #endif
3551 
3552 	kasan_cache_create(s, &size, &s->flags);
3553 #ifdef CONFIG_SLUB_DEBUG
3554 	if (flags & SLAB_RED_ZONE) {
3555 		/*
3556 		 * Add some empty padding so that we can catch
3557 		 * overwrites from earlier objects rather than let
3558 		 * tracking information or the free pointer be
3559 		 * corrupted if a user writes before the start
3560 		 * of the object.
3561 		 */
3562 		size += sizeof(void *);
3563 
3564 		s->red_left_pad = sizeof(void *);
3565 		s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3566 		size += s->red_left_pad;
3567 	}
3568 #endif
3569 
3570 	/*
3571 	 * SLUB stores one object immediately after another beginning from
3572 	 * offset 0. In order to align the objects we have to simply size
3573 	 * each object to conform to the alignment.
3574 	 */
3575 	size = ALIGN(size, s->align);
3576 	s->size = size;
3577 	if (forced_order >= 0)
3578 		order = forced_order;
3579 	else
3580 		order = calculate_order(size);
3581 
3582 	if ((int)order < 0)
3583 		return 0;
3584 
3585 	s->allocflags = 0;
3586 	if (order)
3587 		s->allocflags |= __GFP_COMP;
3588 
3589 	if (s->flags & SLAB_CACHE_DMA)
3590 		s->allocflags |= GFP_DMA;
3591 
3592 	if (s->flags & SLAB_CACHE_DMA32)
3593 		s->allocflags |= GFP_DMA32;
3594 
3595 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
3596 		s->allocflags |= __GFP_RECLAIMABLE;
3597 
3598 	/*
3599 	 * Determine the number of objects per slab
3600 	 */
3601 	s->oo = oo_make(order, size);
3602 	s->min = oo_make(get_order(size), size);
3603 	if (oo_objects(s->oo) > oo_objects(s->max))
3604 		s->max = s->oo;
3605 
3606 	return !!oo_objects(s->oo);
3607 }
3608 
3609 static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
3610 {
3611 	s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
3612 #ifdef CONFIG_SLAB_FREELIST_HARDENED
3613 	s->random = get_random_long();
3614 #endif
3615 
3616 	if (!calculate_sizes(s, -1))
3617 		goto error;
3618 	if (disable_higher_order_debug) {
3619 		/*
3620 		 * Disable debugging flags that store metadata if the min slab
3621 		 * order increased.
3622 		 */
3623 		if (get_order(s->size) > get_order(s->object_size)) {
3624 			s->flags &= ~DEBUG_METADATA_FLAGS;
3625 			s->offset = 0;
3626 			if (!calculate_sizes(s, -1))
3627 				goto error;
3628 		}
3629 	}
3630 
3631 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3632     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3633 	if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
3634 		/* Enable fast mode */
3635 		s->flags |= __CMPXCHG_DOUBLE;
3636 #endif
3637 
3638 	/*
3639 	 * The larger the object size is, the more pages we want on the partial
3640 	 * list to avoid pounding the page allocator excessively.
3641 	 */
3642 	set_min_partial(s, ilog2(s->size) / 2);
3643 
3644 	set_cpu_partial(s);
3645 
3646 #ifdef CONFIG_NUMA
3647 	s->remote_node_defrag_ratio = 1000;
3648 #endif
3649 
3650 	/* Initialize the pre-computed randomized freelist if slab is up */
3651 	if (slab_state >= UP) {
3652 		if (init_cache_random_seq(s))
3653 			goto error;
3654 	}
3655 
3656 	if (!init_kmem_cache_nodes(s))
3657 		goto error;
3658 
3659 	if (alloc_kmem_cache_cpus(s))
3660 		return 0;
3661 
3662 	free_kmem_cache_nodes(s);
3663 error:
3664 	if (flags & SLAB_PANIC)
3665 		panic("Cannot create slab %s size=%u realsize=%u order=%u offset=%u flags=%lx\n",
3666 		      s->name, s->size, s->size,
3667 		      oo_order(s->oo), s->offset, (unsigned long)flags);
3668 	return -EINVAL;
3669 }
3670 
3671 static void list_slab_objects(struct kmem_cache *s, struct page *page,
3672 							const char *text)
3673 {
3674 #ifdef CONFIG_SLUB_DEBUG
3675 	void *addr = page_address(page);
3676 	void *p;
3677 	unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC);
3678 	if (!map)
3679 		return;
3680 	slab_err(s, page, text, s->name);
3681 	slab_lock(page);
3682 
3683 	get_map(s, page, map);
3684 	for_each_object(p, s, addr, page->objects) {
3685 
3686 		if (!test_bit(slab_index(p, s, addr), map)) {
3687 			pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
3688 			print_tracking(s, p);
3689 		}
3690 	}
3691 	slab_unlock(page);
3692 	bitmap_free(map);
3693 #endif
3694 }
3695 
3696 /*
3697  * Attempt to free all partial slabs on a node.
3698  * This is called from __kmem_cache_shutdown(). We must take list_lock
3699  * because sysfs file might still access partial list after the shutdowning.
3700  */
3701 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
3702 {
3703 	LIST_HEAD(discard);
3704 	struct page *page, *h;
3705 
3706 	BUG_ON(irqs_disabled());
3707 	spin_lock_irq(&n->list_lock);
3708 	list_for_each_entry_safe(page, h, &n->partial, lru) {
3709 		if (!page->inuse) {
3710 			remove_partial(n, page);
3711 			list_add(&page->lru, &discard);
3712 		} else {
3713 			list_slab_objects(s, page,
3714 			"Objects remaining in %s on __kmem_cache_shutdown()");
3715 		}
3716 	}
3717 	spin_unlock_irq(&n->list_lock);
3718 
3719 	list_for_each_entry_safe(page, h, &discard, lru)
3720 		discard_slab(s, page);
3721 }
3722 
3723 bool __kmem_cache_empty(struct kmem_cache *s)
3724 {
3725 	int node;
3726 	struct kmem_cache_node *n;
3727 
3728 	for_each_kmem_cache_node(s, node, n)
3729 		if (n->nr_partial || slabs_node(s, node))
3730 			return false;
3731 	return true;
3732 }
3733 
3734 /*
3735  * Release all resources used by a slab cache.
3736  */
3737 int __kmem_cache_shutdown(struct kmem_cache *s)
3738 {
3739 	int node;
3740 	struct kmem_cache_node *n;
3741 
3742 	flush_all(s);
3743 	/* Attempt to free all objects */
3744 	for_each_kmem_cache_node(s, node, n) {
3745 		free_partial(s, n);
3746 		if (n->nr_partial || slabs_node(s, node))
3747 			return 1;
3748 	}
3749 	sysfs_slab_remove(s);
3750 	return 0;
3751 }
3752 
3753 /********************************************************************
3754  *		Kmalloc subsystem
3755  *******************************************************************/
3756 
3757 static int __init setup_slub_min_order(char *str)
3758 {
3759 	get_option(&str, (int *)&slub_min_order);
3760 
3761 	return 1;
3762 }
3763 
3764 __setup("slub_min_order=", setup_slub_min_order);
3765 
3766 static int __init setup_slub_max_order(char *str)
3767 {
3768 	get_option(&str, (int *)&slub_max_order);
3769 	slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
3770 
3771 	return 1;
3772 }
3773 
3774 __setup("slub_max_order=", setup_slub_max_order);
3775 
3776 static int __init setup_slub_min_objects(char *str)
3777 {
3778 	get_option(&str, (int *)&slub_min_objects);
3779 
3780 	return 1;
3781 }
3782 
3783 __setup("slub_min_objects=", setup_slub_min_objects);
3784 
3785 void *__kmalloc(size_t size, gfp_t flags)
3786 {
3787 	struct kmem_cache *s;
3788 	void *ret;
3789 
3790 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3791 		return kmalloc_large(size, flags);
3792 
3793 	s = kmalloc_slab(size, flags);
3794 
3795 	if (unlikely(ZERO_OR_NULL_PTR(s)))
3796 		return s;
3797 
3798 	ret = slab_alloc(s, flags, _RET_IP_);
3799 
3800 	trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
3801 
3802 	ret = kasan_kmalloc(s, ret, size, flags);
3803 
3804 	return ret;
3805 }
3806 EXPORT_SYMBOL(__kmalloc);
3807 
3808 #ifdef CONFIG_NUMA
3809 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3810 {
3811 	struct page *page;
3812 	void *ptr = NULL;
3813 
3814 	flags |= __GFP_COMP;
3815 	page = alloc_pages_node(node, flags, get_order(size));
3816 	if (page)
3817 		ptr = page_address(page);
3818 
3819 	return kmalloc_large_node_hook(ptr, size, flags);
3820 }
3821 
3822 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3823 {
3824 	struct kmem_cache *s;
3825 	void *ret;
3826 
3827 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
3828 		ret = kmalloc_large_node(size, flags, node);
3829 
3830 		trace_kmalloc_node(_RET_IP_, ret,
3831 				   size, PAGE_SIZE << get_order(size),
3832 				   flags, node);
3833 
3834 		return ret;
3835 	}
3836 
3837 	s = kmalloc_slab(size, flags);
3838 
3839 	if (unlikely(ZERO_OR_NULL_PTR(s)))
3840 		return s;
3841 
3842 	ret = slab_alloc_node(s, flags, node, _RET_IP_);
3843 
3844 	trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
3845 
3846 	ret = kasan_kmalloc(s, ret, size, flags);
3847 
3848 	return ret;
3849 }
3850 EXPORT_SYMBOL(__kmalloc_node);
3851 #endif
3852 
3853 #ifdef CONFIG_HARDENED_USERCOPY
3854 /*
3855  * Rejects incorrectly sized objects and objects that are to be copied
3856  * to/from userspace but do not fall entirely within the containing slab
3857  * cache's usercopy region.
3858  *
3859  * Returns NULL if check passes, otherwise const char * to name of cache
3860  * to indicate an error.
3861  */
3862 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
3863 			 bool to_user)
3864 {
3865 	struct kmem_cache *s;
3866 	unsigned int offset;
3867 	size_t object_size;
3868 
3869 	ptr = kasan_reset_tag(ptr);
3870 
3871 	/* Find object and usable object size. */
3872 	s = page->slab_cache;
3873 
3874 	/* Reject impossible pointers. */
3875 	if (ptr < page_address(page))
3876 		usercopy_abort("SLUB object not in SLUB page?!", NULL,
3877 			       to_user, 0, n);
3878 
3879 	/* Find offset within object. */
3880 	offset = (ptr - page_address(page)) % s->size;
3881 
3882 	/* Adjust for redzone and reject if within the redzone. */
3883 	if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3884 		if (offset < s->red_left_pad)
3885 			usercopy_abort("SLUB object in left red zone",
3886 				       s->name, to_user, offset, n);
3887 		offset -= s->red_left_pad;
3888 	}
3889 
3890 	/* Allow address range falling entirely within usercopy region. */
3891 	if (offset >= s->useroffset &&
3892 	    offset - s->useroffset <= s->usersize &&
3893 	    n <= s->useroffset - offset + s->usersize)
3894 		return;
3895 
3896 	/*
3897 	 * If the copy is still within the allocated object, produce
3898 	 * a warning instead of rejecting the copy. This is intended
3899 	 * to be a temporary method to find any missing usercopy
3900 	 * whitelists.
3901 	 */
3902 	object_size = slab_ksize(s);
3903 	if (usercopy_fallback &&
3904 	    offset <= object_size && n <= object_size - offset) {
3905 		usercopy_warn("SLUB object", s->name, to_user, offset, n);
3906 		return;
3907 	}
3908 
3909 	usercopy_abort("SLUB object", s->name, to_user, offset, n);
3910 }
3911 #endif /* CONFIG_HARDENED_USERCOPY */
3912 
3913 static size_t __ksize(const void *object)
3914 {
3915 	struct page *page;
3916 
3917 	if (unlikely(object == ZERO_SIZE_PTR))
3918 		return 0;
3919 
3920 	page = virt_to_head_page(object);
3921 
3922 	if (unlikely(!PageSlab(page))) {
3923 		WARN_ON(!PageCompound(page));
3924 		return PAGE_SIZE << compound_order(page);
3925 	}
3926 
3927 	return slab_ksize(page->slab_cache);
3928 }
3929 
3930 size_t ksize(const void *object)
3931 {
3932 	size_t size = __ksize(object);
3933 	/* We assume that ksize callers could use whole allocated area,
3934 	 * so we need to unpoison this area.
3935 	 */
3936 	kasan_unpoison_shadow(object, size);
3937 	return size;
3938 }
3939 EXPORT_SYMBOL(ksize);
3940 
3941 void kfree(const void *x)
3942 {
3943 	struct page *page;
3944 	void *object = (void *)x;
3945 
3946 	trace_kfree(_RET_IP_, x);
3947 
3948 	if (unlikely(ZERO_OR_NULL_PTR(x)))
3949 		return;
3950 
3951 	page = virt_to_head_page(x);
3952 	if (unlikely(!PageSlab(page))) {
3953 		BUG_ON(!PageCompound(page));
3954 		kfree_hook(object);
3955 		__free_pages(page, compound_order(page));
3956 		return;
3957 	}
3958 	slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
3959 }
3960 EXPORT_SYMBOL(kfree);
3961 
3962 #define SHRINK_PROMOTE_MAX 32
3963 
3964 /*
3965  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3966  * up most to the head of the partial lists. New allocations will then
3967  * fill those up and thus they can be removed from the partial lists.
3968  *
3969  * The slabs with the least items are placed last. This results in them
3970  * being allocated from last increasing the chance that the last objects
3971  * are freed in them.
3972  */
3973 int __kmem_cache_shrink(struct kmem_cache *s)
3974 {
3975 	int node;
3976 	int i;
3977 	struct kmem_cache_node *n;
3978 	struct page *page;
3979 	struct page *t;
3980 	struct list_head discard;
3981 	struct list_head promote[SHRINK_PROMOTE_MAX];
3982 	unsigned long flags;
3983 	int ret = 0;
3984 
3985 	flush_all(s);
3986 	for_each_kmem_cache_node(s, node, n) {
3987 		INIT_LIST_HEAD(&discard);
3988 		for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3989 			INIT_LIST_HEAD(promote + i);
3990 
3991 		spin_lock_irqsave(&n->list_lock, flags);
3992 
3993 		/*
3994 		 * Build lists of slabs to discard or promote.
3995 		 *
3996 		 * Note that concurrent frees may occur while we hold the
3997 		 * list_lock. page->inuse here is the upper limit.
3998 		 */
3999 		list_for_each_entry_safe(page, t, &n->partial, lru) {
4000 			int free = page->objects - page->inuse;
4001 
4002 			/* Do not reread page->inuse */
4003 			barrier();
4004 
4005 			/* We do not keep full slabs on the list */
4006 			BUG_ON(free <= 0);
4007 
4008 			if (free == page->objects) {
4009 				list_move(&page->lru, &discard);
4010 				n->nr_partial--;
4011 			} else if (free <= SHRINK_PROMOTE_MAX)
4012 				list_move(&page->lru, promote + free - 1);
4013 		}
4014 
4015 		/*
4016 		 * Promote the slabs filled up most to the head of the
4017 		 * partial list.
4018 		 */
4019 		for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4020 			list_splice(promote + i, &n->partial);
4021 
4022 		spin_unlock_irqrestore(&n->list_lock, flags);
4023 
4024 		/* Release empty slabs */
4025 		list_for_each_entry_safe(page, t, &discard, lru)
4026 			discard_slab(s, page);
4027 
4028 		if (slabs_node(s, node))
4029 			ret = 1;
4030 	}
4031 
4032 	return ret;
4033 }
4034 
4035 #ifdef CONFIG_MEMCG
4036 static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s)
4037 {
4038 	/*
4039 	 * Called with all the locks held after a sched RCU grace period.
4040 	 * Even if @s becomes empty after shrinking, we can't know that @s
4041 	 * doesn't have allocations already in-flight and thus can't
4042 	 * destroy @s until the associated memcg is released.
4043 	 *
4044 	 * However, let's remove the sysfs files for empty caches here.
4045 	 * Each cache has a lot of interface files which aren't
4046 	 * particularly useful for empty draining caches; otherwise, we can
4047 	 * easily end up with millions of unnecessary sysfs files on
4048 	 * systems which have a lot of memory and transient cgroups.
4049 	 */
4050 	if (!__kmem_cache_shrink(s))
4051 		sysfs_slab_remove(s);
4052 }
4053 
4054 void __kmemcg_cache_deactivate(struct kmem_cache *s)
4055 {
4056 	/*
4057 	 * Disable empty slabs caching. Used to avoid pinning offline
4058 	 * memory cgroups by kmem pages that can be freed.
4059 	 */
4060 	slub_set_cpu_partial(s, 0);
4061 	s->min_partial = 0;
4062 
4063 	/*
4064 	 * s->cpu_partial is checked locklessly (see put_cpu_partial), so
4065 	 * we have to make sure the change is visible before shrinking.
4066 	 */
4067 	slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu);
4068 }
4069 #endif
4070 
4071 static int slab_mem_going_offline_callback(void *arg)
4072 {
4073 	struct kmem_cache *s;
4074 
4075 	mutex_lock(&slab_mutex);
4076 	list_for_each_entry(s, &slab_caches, list)
4077 		__kmem_cache_shrink(s);
4078 	mutex_unlock(&slab_mutex);
4079 
4080 	return 0;
4081 }
4082 
4083 static void slab_mem_offline_callback(void *arg)
4084 {
4085 	struct kmem_cache_node *n;
4086 	struct kmem_cache *s;
4087 	struct memory_notify *marg = arg;
4088 	int offline_node;
4089 
4090 	offline_node = marg->status_change_nid_normal;
4091 
4092 	/*
4093 	 * If the node still has available memory. we need kmem_cache_node
4094 	 * for it yet.
4095 	 */
4096 	if (offline_node < 0)
4097 		return;
4098 
4099 	mutex_lock(&slab_mutex);
4100 	list_for_each_entry(s, &slab_caches, list) {
4101 		n = get_node(s, offline_node);
4102 		if (n) {
4103 			/*
4104 			 * if n->nr_slabs > 0, slabs still exist on the node
4105 			 * that is going down. We were unable to free them,
4106 			 * and offline_pages() function shouldn't call this
4107 			 * callback. So, we must fail.
4108 			 */
4109 			BUG_ON(slabs_node(s, offline_node));
4110 
4111 			s->node[offline_node] = NULL;
4112 			kmem_cache_free(kmem_cache_node, n);
4113 		}
4114 	}
4115 	mutex_unlock(&slab_mutex);
4116 }
4117 
4118 static int slab_mem_going_online_callback(void *arg)
4119 {
4120 	struct kmem_cache_node *n;
4121 	struct kmem_cache *s;
4122 	struct memory_notify *marg = arg;
4123 	int nid = marg->status_change_nid_normal;
4124 	int ret = 0;
4125 
4126 	/*
4127 	 * If the node's memory is already available, then kmem_cache_node is
4128 	 * already created. Nothing to do.
4129 	 */
4130 	if (nid < 0)
4131 		return 0;
4132 
4133 	/*
4134 	 * We are bringing a node online. No memory is available yet. We must
4135 	 * allocate a kmem_cache_node structure in order to bring the node
4136 	 * online.
4137 	 */
4138 	mutex_lock(&slab_mutex);
4139 	list_for_each_entry(s, &slab_caches, list) {
4140 		/*
4141 		 * XXX: kmem_cache_alloc_node will fallback to other nodes
4142 		 *      since memory is not yet available from the node that
4143 		 *      is brought up.
4144 		 */
4145 		n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4146 		if (!n) {
4147 			ret = -ENOMEM;
4148 			goto out;
4149 		}
4150 		init_kmem_cache_node(n);
4151 		s->node[nid] = n;
4152 	}
4153 out:
4154 	mutex_unlock(&slab_mutex);
4155 	return ret;
4156 }
4157 
4158 static int slab_memory_callback(struct notifier_block *self,
4159 				unsigned long action, void *arg)
4160 {
4161 	int ret = 0;
4162 
4163 	switch (action) {
4164 	case MEM_GOING_ONLINE:
4165 		ret = slab_mem_going_online_callback(arg);
4166 		break;
4167 	case MEM_GOING_OFFLINE:
4168 		ret = slab_mem_going_offline_callback(arg);
4169 		break;
4170 	case MEM_OFFLINE:
4171 	case MEM_CANCEL_ONLINE:
4172 		slab_mem_offline_callback(arg);
4173 		break;
4174 	case MEM_ONLINE:
4175 	case MEM_CANCEL_OFFLINE:
4176 		break;
4177 	}
4178 	if (ret)
4179 		ret = notifier_from_errno(ret);
4180 	else
4181 		ret = NOTIFY_OK;
4182 	return ret;
4183 }
4184 
4185 static struct notifier_block slab_memory_callback_nb = {
4186 	.notifier_call = slab_memory_callback,
4187 	.priority = SLAB_CALLBACK_PRI,
4188 };
4189 
4190 /********************************************************************
4191  *			Basic setup of slabs
4192  *******************************************************************/
4193 
4194 /*
4195  * Used for early kmem_cache structures that were allocated using
4196  * the page allocator. Allocate them properly then fix up the pointers
4197  * that may be pointing to the wrong kmem_cache structure.
4198  */
4199 
4200 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4201 {
4202 	int node;
4203 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4204 	struct kmem_cache_node *n;
4205 
4206 	memcpy(s, static_cache, kmem_cache->object_size);
4207 
4208 	/*
4209 	 * This runs very early, and only the boot processor is supposed to be
4210 	 * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4211 	 * IPIs around.
4212 	 */
4213 	__flush_cpu_slab(s, smp_processor_id());
4214 	for_each_kmem_cache_node(s, node, n) {
4215 		struct page *p;
4216 
4217 		list_for_each_entry(p, &n->partial, lru)
4218 			p->slab_cache = s;
4219 
4220 #ifdef CONFIG_SLUB_DEBUG
4221 		list_for_each_entry(p, &n->full, lru)
4222 			p->slab_cache = s;
4223 #endif
4224 	}
4225 	slab_init_memcg_params(s);
4226 	list_add(&s->list, &slab_caches);
4227 	memcg_link_cache(s);
4228 	return s;
4229 }
4230 
4231 void __init kmem_cache_init(void)
4232 {
4233 	static __initdata struct kmem_cache boot_kmem_cache,
4234 		boot_kmem_cache_node;
4235 
4236 	if (debug_guardpage_minorder())
4237 		slub_max_order = 0;
4238 
4239 	kmem_cache_node = &boot_kmem_cache_node;
4240 	kmem_cache = &boot_kmem_cache;
4241 
4242 	create_boot_cache(kmem_cache_node, "kmem_cache_node",
4243 		sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
4244 
4245 	register_hotmemory_notifier(&slab_memory_callback_nb);
4246 
4247 	/* Able to allocate the per node structures */
4248 	slab_state = PARTIAL;
4249 
4250 	create_boot_cache(kmem_cache, "kmem_cache",
4251 			offsetof(struct kmem_cache, node) +
4252 				nr_node_ids * sizeof(struct kmem_cache_node *),
4253 		       SLAB_HWCACHE_ALIGN, 0, 0);
4254 
4255 	kmem_cache = bootstrap(&boot_kmem_cache);
4256 	kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4257 
4258 	/* Now we can use the kmem_cache to allocate kmalloc slabs */
4259 	setup_kmalloc_cache_index_table();
4260 	create_kmalloc_caches(0);
4261 
4262 	/* Setup random freelists for each cache */
4263 	init_freelist_randomization();
4264 
4265 	cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4266 				  slub_cpu_dead);
4267 
4268 	pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
4269 		cache_line_size(),
4270 		slub_min_order, slub_max_order, slub_min_objects,
4271 		nr_cpu_ids, nr_node_ids);
4272 }
4273 
4274 void __init kmem_cache_init_late(void)
4275 {
4276 }
4277 
4278 struct kmem_cache *
4279 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
4280 		   slab_flags_t flags, void (*ctor)(void *))
4281 {
4282 	struct kmem_cache *s, *c;
4283 
4284 	s = find_mergeable(size, align, flags, name, ctor);
4285 	if (s) {
4286 		s->refcount++;
4287 
4288 		/*
4289 		 * Adjust the object sizes so that we clear
4290 		 * the complete object on kzalloc.
4291 		 */
4292 		s->object_size = max(s->object_size, size);
4293 		s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
4294 
4295 		for_each_memcg_cache(c, s) {
4296 			c->object_size = s->object_size;
4297 			c->inuse = max(c->inuse, ALIGN(size, sizeof(void *)));
4298 		}
4299 
4300 		if (sysfs_slab_alias(s, name)) {
4301 			s->refcount--;
4302 			s = NULL;
4303 		}
4304 	}
4305 
4306 	return s;
4307 }
4308 
4309 int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
4310 {
4311 	int err;
4312 
4313 	err = kmem_cache_open(s, flags);
4314 	if (err)
4315 		return err;
4316 
4317 	/* Mutex is not taken during early boot */
4318 	if (slab_state <= UP)
4319 		return 0;
4320 
4321 	memcg_propagate_slab_attrs(s);
4322 	err = sysfs_slab_add(s);
4323 	if (err)
4324 		__kmem_cache_release(s);
4325 
4326 	return err;
4327 }
4328 
4329 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4330 {
4331 	struct kmem_cache *s;
4332 	void *ret;
4333 
4334 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4335 		return kmalloc_large(size, gfpflags);
4336 
4337 	s = kmalloc_slab(size, gfpflags);
4338 
4339 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4340 		return s;
4341 
4342 	ret = slab_alloc(s, gfpflags, caller);
4343 
4344 	/* Honor the call site pointer we received. */
4345 	trace_kmalloc(caller, ret, size, s->size, gfpflags);
4346 
4347 	return ret;
4348 }
4349 
4350 #ifdef CONFIG_NUMA
4351 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4352 					int node, unsigned long caller)
4353 {
4354 	struct kmem_cache *s;
4355 	void *ret;
4356 
4357 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4358 		ret = kmalloc_large_node(size, gfpflags, node);
4359 
4360 		trace_kmalloc_node(caller, ret,
4361 				   size, PAGE_SIZE << get_order(size),
4362 				   gfpflags, node);
4363 
4364 		return ret;
4365 	}
4366 
4367 	s = kmalloc_slab(size, gfpflags);
4368 
4369 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4370 		return s;
4371 
4372 	ret = slab_alloc_node(s, gfpflags, node, caller);
4373 
4374 	/* Honor the call site pointer we received. */
4375 	trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4376 
4377 	return ret;
4378 }
4379 #endif
4380 
4381 #ifdef CONFIG_SYSFS
4382 static int count_inuse(struct page *page)
4383 {
4384 	return page->inuse;
4385 }
4386 
4387 static int count_total(struct page *page)
4388 {
4389 	return page->objects;
4390 }
4391 #endif
4392 
4393 #ifdef CONFIG_SLUB_DEBUG
4394 static int validate_slab(struct kmem_cache *s, struct page *page,
4395 						unsigned long *map)
4396 {
4397 	void *p;
4398 	void *addr = page_address(page);
4399 
4400 	if (!check_slab(s, page) ||
4401 			!on_freelist(s, page, NULL))
4402 		return 0;
4403 
4404 	/* Now we know that a valid freelist exists */
4405 	bitmap_zero(map, page->objects);
4406 
4407 	get_map(s, page, map);
4408 	for_each_object(p, s, addr, page->objects) {
4409 		if (test_bit(slab_index(p, s, addr), map))
4410 			if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4411 				return 0;
4412 	}
4413 
4414 	for_each_object(p, s, addr, page->objects)
4415 		if (!test_bit(slab_index(p, s, addr), map))
4416 			if (!check_object(s, page, p, SLUB_RED_ACTIVE))
4417 				return 0;
4418 	return 1;
4419 }
4420 
4421 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4422 						unsigned long *map)
4423 {
4424 	slab_lock(page);
4425 	validate_slab(s, page, map);
4426 	slab_unlock(page);
4427 }
4428 
4429 static int validate_slab_node(struct kmem_cache *s,
4430 		struct kmem_cache_node *n, unsigned long *map)
4431 {
4432 	unsigned long count = 0;
4433 	struct page *page;
4434 	unsigned long flags;
4435 
4436 	spin_lock_irqsave(&n->list_lock, flags);
4437 
4438 	list_for_each_entry(page, &n->partial, lru) {
4439 		validate_slab_slab(s, page, map);
4440 		count++;
4441 	}
4442 	if (count != n->nr_partial)
4443 		pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4444 		       s->name, count, n->nr_partial);
4445 
4446 	if (!(s->flags & SLAB_STORE_USER))
4447 		goto out;
4448 
4449 	list_for_each_entry(page, &n->full, lru) {
4450 		validate_slab_slab(s, page, map);
4451 		count++;
4452 	}
4453 	if (count != atomic_long_read(&n->nr_slabs))
4454 		pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4455 		       s->name, count, atomic_long_read(&n->nr_slabs));
4456 
4457 out:
4458 	spin_unlock_irqrestore(&n->list_lock, flags);
4459 	return count;
4460 }
4461 
4462 static long validate_slab_cache(struct kmem_cache *s)
4463 {
4464 	int node;
4465 	unsigned long count = 0;
4466 	struct kmem_cache_node *n;
4467 	unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
4468 
4469 	if (!map)
4470 		return -ENOMEM;
4471 
4472 	flush_all(s);
4473 	for_each_kmem_cache_node(s, node, n)
4474 		count += validate_slab_node(s, n, map);
4475 	bitmap_free(map);
4476 	return count;
4477 }
4478 /*
4479  * Generate lists of code addresses where slabcache objects are allocated
4480  * and freed.
4481  */
4482 
4483 struct location {
4484 	unsigned long count;
4485 	unsigned long addr;
4486 	long long sum_time;
4487 	long min_time;
4488 	long max_time;
4489 	long min_pid;
4490 	long max_pid;
4491 	DECLARE_BITMAP(cpus, NR_CPUS);
4492 	nodemask_t nodes;
4493 };
4494 
4495 struct loc_track {
4496 	unsigned long max;
4497 	unsigned long count;
4498 	struct location *loc;
4499 };
4500 
4501 static void free_loc_track(struct loc_track *t)
4502 {
4503 	if (t->max)
4504 		free_pages((unsigned long)t->loc,
4505 			get_order(sizeof(struct location) * t->max));
4506 }
4507 
4508 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
4509 {
4510 	struct location *l;
4511 	int order;
4512 
4513 	order = get_order(sizeof(struct location) * max);
4514 
4515 	l = (void *)__get_free_pages(flags, order);
4516 	if (!l)
4517 		return 0;
4518 
4519 	if (t->count) {
4520 		memcpy(l, t->loc, sizeof(struct location) * t->count);
4521 		free_loc_track(t);
4522 	}
4523 	t->max = max;
4524 	t->loc = l;
4525 	return 1;
4526 }
4527 
4528 static int add_location(struct loc_track *t, struct kmem_cache *s,
4529 				const struct track *track)
4530 {
4531 	long start, end, pos;
4532 	struct location *l;
4533 	unsigned long caddr;
4534 	unsigned long age = jiffies - track->when;
4535 
4536 	start = -1;
4537 	end = t->count;
4538 
4539 	for ( ; ; ) {
4540 		pos = start + (end - start + 1) / 2;
4541 
4542 		/*
4543 		 * There is nothing at "end". If we end up there
4544 		 * we need to add something to before end.
4545 		 */
4546 		if (pos == end)
4547 			break;
4548 
4549 		caddr = t->loc[pos].addr;
4550 		if (track->addr == caddr) {
4551 
4552 			l = &t->loc[pos];
4553 			l->count++;
4554 			if (track->when) {
4555 				l->sum_time += age;
4556 				if (age < l->min_time)
4557 					l->min_time = age;
4558 				if (age > l->max_time)
4559 					l->max_time = age;
4560 
4561 				if (track->pid < l->min_pid)
4562 					l->min_pid = track->pid;
4563 				if (track->pid > l->max_pid)
4564 					l->max_pid = track->pid;
4565 
4566 				cpumask_set_cpu(track->cpu,
4567 						to_cpumask(l->cpus));
4568 			}
4569 			node_set(page_to_nid(virt_to_page(track)), l->nodes);
4570 			return 1;
4571 		}
4572 
4573 		if (track->addr < caddr)
4574 			end = pos;
4575 		else
4576 			start = pos;
4577 	}
4578 
4579 	/*
4580 	 * Not found. Insert new tracking element.
4581 	 */
4582 	if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
4583 		return 0;
4584 
4585 	l = t->loc + pos;
4586 	if (pos < t->count)
4587 		memmove(l + 1, l,
4588 			(t->count - pos) * sizeof(struct location));
4589 	t->count++;
4590 	l->count = 1;
4591 	l->addr = track->addr;
4592 	l->sum_time = age;
4593 	l->min_time = age;
4594 	l->max_time = age;
4595 	l->min_pid = track->pid;
4596 	l->max_pid = track->pid;
4597 	cpumask_clear(to_cpumask(l->cpus));
4598 	cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
4599 	nodes_clear(l->nodes);
4600 	node_set(page_to_nid(virt_to_page(track)), l->nodes);
4601 	return 1;
4602 }
4603 
4604 static void process_slab(struct loc_track *t, struct kmem_cache *s,
4605 		struct page *page, enum track_item alloc,
4606 		unsigned long *map)
4607 {
4608 	void *addr = page_address(page);
4609 	void *p;
4610 
4611 	bitmap_zero(map, page->objects);
4612 	get_map(s, page, map);
4613 
4614 	for_each_object(p, s, addr, page->objects)
4615 		if (!test_bit(slab_index(p, s, addr), map))
4616 			add_location(t, s, get_track(s, p, alloc));
4617 }
4618 
4619 static int list_locations(struct kmem_cache *s, char *buf,
4620 					enum track_item alloc)
4621 {
4622 	int len = 0;
4623 	unsigned long i;
4624 	struct loc_track t = { 0, 0, NULL };
4625 	int node;
4626 	struct kmem_cache_node *n;
4627 	unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
4628 
4629 	if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4630 				     GFP_KERNEL)) {
4631 		bitmap_free(map);
4632 		return sprintf(buf, "Out of memory\n");
4633 	}
4634 	/* Push back cpu slabs */
4635 	flush_all(s);
4636 
4637 	for_each_kmem_cache_node(s, node, n) {
4638 		unsigned long flags;
4639 		struct page *page;
4640 
4641 		if (!atomic_long_read(&n->nr_slabs))
4642 			continue;
4643 
4644 		spin_lock_irqsave(&n->list_lock, flags);
4645 		list_for_each_entry(page, &n->partial, lru)
4646 			process_slab(&t, s, page, alloc, map);
4647 		list_for_each_entry(page, &n->full, lru)
4648 			process_slab(&t, s, page, alloc, map);
4649 		spin_unlock_irqrestore(&n->list_lock, flags);
4650 	}
4651 
4652 	for (i = 0; i < t.count; i++) {
4653 		struct location *l = &t.loc[i];
4654 
4655 		if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
4656 			break;
4657 		len += sprintf(buf + len, "%7ld ", l->count);
4658 
4659 		if (l->addr)
4660 			len += sprintf(buf + len, "%pS", (void *)l->addr);
4661 		else
4662 			len += sprintf(buf + len, "<not-available>");
4663 
4664 		if (l->sum_time != l->min_time) {
4665 			len += sprintf(buf + len, " age=%ld/%ld/%ld",
4666 				l->min_time,
4667 				(long)div_u64(l->sum_time, l->count),
4668 				l->max_time);
4669 		} else
4670 			len += sprintf(buf + len, " age=%ld",
4671 				l->min_time);
4672 
4673 		if (l->min_pid != l->max_pid)
4674 			len += sprintf(buf + len, " pid=%ld-%ld",
4675 				l->min_pid, l->max_pid);
4676 		else
4677 			len += sprintf(buf + len, " pid=%ld",
4678 				l->min_pid);
4679 
4680 		if (num_online_cpus() > 1 &&
4681 				!cpumask_empty(to_cpumask(l->cpus)) &&
4682 				len < PAGE_SIZE - 60)
4683 			len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4684 					 " cpus=%*pbl",
4685 					 cpumask_pr_args(to_cpumask(l->cpus)));
4686 
4687 		if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
4688 				len < PAGE_SIZE - 60)
4689 			len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4690 					 " nodes=%*pbl",
4691 					 nodemask_pr_args(&l->nodes));
4692 
4693 		len += sprintf(buf + len, "\n");
4694 	}
4695 
4696 	free_loc_track(&t);
4697 	bitmap_free(map);
4698 	if (!t.count)
4699 		len += sprintf(buf, "No data\n");
4700 	return len;
4701 }
4702 #endif
4703 
4704 #ifdef SLUB_RESILIENCY_TEST
4705 static void __init resiliency_test(void)
4706 {
4707 	u8 *p;
4708 	int type = KMALLOC_NORMAL;
4709 
4710 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
4711 
4712 	pr_err("SLUB resiliency testing\n");
4713 	pr_err("-----------------------\n");
4714 	pr_err("A. Corruption after allocation\n");
4715 
4716 	p = kzalloc(16, GFP_KERNEL);
4717 	p[16] = 0x12;
4718 	pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4719 	       p + 16);
4720 
4721 	validate_slab_cache(kmalloc_caches[type][4]);
4722 
4723 	/* Hmmm... The next two are dangerous */
4724 	p = kzalloc(32, GFP_KERNEL);
4725 	p[32 + sizeof(void *)] = 0x34;
4726 	pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4727 	       p);
4728 	pr_err("If allocated object is overwritten then not detectable\n\n");
4729 
4730 	validate_slab_cache(kmalloc_caches[type][5]);
4731 	p = kzalloc(64, GFP_KERNEL);
4732 	p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4733 	*p = 0x56;
4734 	pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4735 	       p);
4736 	pr_err("If allocated object is overwritten then not detectable\n\n");
4737 	validate_slab_cache(kmalloc_caches[type][6]);
4738 
4739 	pr_err("\nB. Corruption after free\n");
4740 	p = kzalloc(128, GFP_KERNEL);
4741 	kfree(p);
4742 	*p = 0x78;
4743 	pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4744 	validate_slab_cache(kmalloc_caches[type][7]);
4745 
4746 	p = kzalloc(256, GFP_KERNEL);
4747 	kfree(p);
4748 	p[50] = 0x9a;
4749 	pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
4750 	validate_slab_cache(kmalloc_caches[type][8]);
4751 
4752 	p = kzalloc(512, GFP_KERNEL);
4753 	kfree(p);
4754 	p[512] = 0xab;
4755 	pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4756 	validate_slab_cache(kmalloc_caches[type][9]);
4757 }
4758 #else
4759 #ifdef CONFIG_SYSFS
4760 static void resiliency_test(void) {};
4761 #endif
4762 #endif
4763 
4764 #ifdef CONFIG_SYSFS
4765 enum slab_stat_type {
4766 	SL_ALL,			/* All slabs */
4767 	SL_PARTIAL,		/* Only partially allocated slabs */
4768 	SL_CPU,			/* Only slabs used for cpu caches */
4769 	SL_OBJECTS,		/* Determine allocated objects not slabs */
4770 	SL_TOTAL		/* Determine object capacity not slabs */
4771 };
4772 
4773 #define SO_ALL		(1 << SL_ALL)
4774 #define SO_PARTIAL	(1 << SL_PARTIAL)
4775 #define SO_CPU		(1 << SL_CPU)
4776 #define SO_OBJECTS	(1 << SL_OBJECTS)
4777 #define SO_TOTAL	(1 << SL_TOTAL)
4778 
4779 #ifdef CONFIG_MEMCG
4780 static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4781 
4782 static int __init setup_slub_memcg_sysfs(char *str)
4783 {
4784 	int v;
4785 
4786 	if (get_option(&str, &v) > 0)
4787 		memcg_sysfs_enabled = v;
4788 
4789 	return 1;
4790 }
4791 
4792 __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4793 #endif
4794 
4795 static ssize_t show_slab_objects(struct kmem_cache *s,
4796 			    char *buf, unsigned long flags)
4797 {
4798 	unsigned long total = 0;
4799 	int node;
4800 	int x;
4801 	unsigned long *nodes;
4802 
4803 	nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
4804 	if (!nodes)
4805 		return -ENOMEM;
4806 
4807 	if (flags & SO_CPU) {
4808 		int cpu;
4809 
4810 		for_each_possible_cpu(cpu) {
4811 			struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4812 							       cpu);
4813 			int node;
4814 			struct page *page;
4815 
4816 			page = READ_ONCE(c->page);
4817 			if (!page)
4818 				continue;
4819 
4820 			node = page_to_nid(page);
4821 			if (flags & SO_TOTAL)
4822 				x = page->objects;
4823 			else if (flags & SO_OBJECTS)
4824 				x = page->inuse;
4825 			else
4826 				x = 1;
4827 
4828 			total += x;
4829 			nodes[node] += x;
4830 
4831 			page = slub_percpu_partial_read_once(c);
4832 			if (page) {
4833 				node = page_to_nid(page);
4834 				if (flags & SO_TOTAL)
4835 					WARN_ON_ONCE(1);
4836 				else if (flags & SO_OBJECTS)
4837 					WARN_ON_ONCE(1);
4838 				else
4839 					x = page->pages;
4840 				total += x;
4841 				nodes[node] += x;
4842 			}
4843 		}
4844 	}
4845 
4846 	get_online_mems();
4847 #ifdef CONFIG_SLUB_DEBUG
4848 	if (flags & SO_ALL) {
4849 		struct kmem_cache_node *n;
4850 
4851 		for_each_kmem_cache_node(s, node, n) {
4852 
4853 			if (flags & SO_TOTAL)
4854 				x = atomic_long_read(&n->total_objects);
4855 			else if (flags & SO_OBJECTS)
4856 				x = atomic_long_read(&n->total_objects) -
4857 					count_partial(n, count_free);
4858 			else
4859 				x = atomic_long_read(&n->nr_slabs);
4860 			total += x;
4861 			nodes[node] += x;
4862 		}
4863 
4864 	} else
4865 #endif
4866 	if (flags & SO_PARTIAL) {
4867 		struct kmem_cache_node *n;
4868 
4869 		for_each_kmem_cache_node(s, node, n) {
4870 			if (flags & SO_TOTAL)
4871 				x = count_partial(n, count_total);
4872 			else if (flags & SO_OBJECTS)
4873 				x = count_partial(n, count_inuse);
4874 			else
4875 				x = n->nr_partial;
4876 			total += x;
4877 			nodes[node] += x;
4878 		}
4879 	}
4880 	x = sprintf(buf, "%lu", total);
4881 #ifdef CONFIG_NUMA
4882 	for (node = 0; node < nr_node_ids; node++)
4883 		if (nodes[node])
4884 			x += sprintf(buf + x, " N%d=%lu",
4885 					node, nodes[node]);
4886 #endif
4887 	put_online_mems();
4888 	kfree(nodes);
4889 	return x + sprintf(buf + x, "\n");
4890 }
4891 
4892 #ifdef CONFIG_SLUB_DEBUG
4893 static int any_slab_objects(struct kmem_cache *s)
4894 {
4895 	int node;
4896 	struct kmem_cache_node *n;
4897 
4898 	for_each_kmem_cache_node(s, node, n)
4899 		if (atomic_long_read(&n->total_objects))
4900 			return 1;
4901 
4902 	return 0;
4903 }
4904 #endif
4905 
4906 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4907 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4908 
4909 struct slab_attribute {
4910 	struct attribute attr;
4911 	ssize_t (*show)(struct kmem_cache *s, char *buf);
4912 	ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4913 };
4914 
4915 #define SLAB_ATTR_RO(_name) \
4916 	static struct slab_attribute _name##_attr = \
4917 	__ATTR(_name, 0400, _name##_show, NULL)
4918 
4919 #define SLAB_ATTR(_name) \
4920 	static struct slab_attribute _name##_attr =  \
4921 	__ATTR(_name, 0600, _name##_show, _name##_store)
4922 
4923 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4924 {
4925 	return sprintf(buf, "%u\n", s->size);
4926 }
4927 SLAB_ATTR_RO(slab_size);
4928 
4929 static ssize_t align_show(struct kmem_cache *s, char *buf)
4930 {
4931 	return sprintf(buf, "%u\n", s->align);
4932 }
4933 SLAB_ATTR_RO(align);
4934 
4935 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4936 {
4937 	return sprintf(buf, "%u\n", s->object_size);
4938 }
4939 SLAB_ATTR_RO(object_size);
4940 
4941 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4942 {
4943 	return sprintf(buf, "%u\n", oo_objects(s->oo));
4944 }
4945 SLAB_ATTR_RO(objs_per_slab);
4946 
4947 static ssize_t order_store(struct kmem_cache *s,
4948 				const char *buf, size_t length)
4949 {
4950 	unsigned int order;
4951 	int err;
4952 
4953 	err = kstrtouint(buf, 10, &order);
4954 	if (err)
4955 		return err;
4956 
4957 	if (order > slub_max_order || order < slub_min_order)
4958 		return -EINVAL;
4959 
4960 	calculate_sizes(s, order);
4961 	return length;
4962 }
4963 
4964 static ssize_t order_show(struct kmem_cache *s, char *buf)
4965 {
4966 	return sprintf(buf, "%u\n", oo_order(s->oo));
4967 }
4968 SLAB_ATTR(order);
4969 
4970 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4971 {
4972 	return sprintf(buf, "%lu\n", s->min_partial);
4973 }
4974 
4975 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4976 				 size_t length)
4977 {
4978 	unsigned long min;
4979 	int err;
4980 
4981 	err = kstrtoul(buf, 10, &min);
4982 	if (err)
4983 		return err;
4984 
4985 	set_min_partial(s, min);
4986 	return length;
4987 }
4988 SLAB_ATTR(min_partial);
4989 
4990 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4991 {
4992 	return sprintf(buf, "%u\n", slub_cpu_partial(s));
4993 }
4994 
4995 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4996 				 size_t length)
4997 {
4998 	unsigned int objects;
4999 	int err;
5000 
5001 	err = kstrtouint(buf, 10, &objects);
5002 	if (err)
5003 		return err;
5004 	if (objects && !kmem_cache_has_cpu_partial(s))
5005 		return -EINVAL;
5006 
5007 	slub_set_cpu_partial(s, objects);
5008 	flush_all(s);
5009 	return length;
5010 }
5011 SLAB_ATTR(cpu_partial);
5012 
5013 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5014 {
5015 	if (!s->ctor)
5016 		return 0;
5017 	return sprintf(buf, "%pS\n", s->ctor);
5018 }
5019 SLAB_ATTR_RO(ctor);
5020 
5021 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5022 {
5023 	return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5024 }
5025 SLAB_ATTR_RO(aliases);
5026 
5027 static ssize_t partial_show(struct kmem_cache *s, char *buf)
5028 {
5029 	return show_slab_objects(s, buf, SO_PARTIAL);
5030 }
5031 SLAB_ATTR_RO(partial);
5032 
5033 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5034 {
5035 	return show_slab_objects(s, buf, SO_CPU);
5036 }
5037 SLAB_ATTR_RO(cpu_slabs);
5038 
5039 static ssize_t objects_show(struct kmem_cache *s, char *buf)
5040 {
5041 	return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5042 }
5043 SLAB_ATTR_RO(objects);
5044 
5045 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5046 {
5047 	return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5048 }
5049 SLAB_ATTR_RO(objects_partial);
5050 
5051 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5052 {
5053 	int objects = 0;
5054 	int pages = 0;
5055 	int cpu;
5056 	int len;
5057 
5058 	for_each_online_cpu(cpu) {
5059 		struct page *page;
5060 
5061 		page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5062 
5063 		if (page) {
5064 			pages += page->pages;
5065 			objects += page->pobjects;
5066 		}
5067 	}
5068 
5069 	len = sprintf(buf, "%d(%d)", objects, pages);
5070 
5071 #ifdef CONFIG_SMP
5072 	for_each_online_cpu(cpu) {
5073 		struct page *page;
5074 
5075 		page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5076 
5077 		if (page && len < PAGE_SIZE - 20)
5078 			len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5079 				page->pobjects, page->pages);
5080 	}
5081 #endif
5082 	return len + sprintf(buf + len, "\n");
5083 }
5084 SLAB_ATTR_RO(slabs_cpu_partial);
5085 
5086 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5087 {
5088 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5089 }
5090 
5091 static ssize_t reclaim_account_store(struct kmem_cache *s,
5092 				const char *buf, size_t length)
5093 {
5094 	s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5095 	if (buf[0] == '1')
5096 		s->flags |= SLAB_RECLAIM_ACCOUNT;
5097 	return length;
5098 }
5099 SLAB_ATTR(reclaim_account);
5100 
5101 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5102 {
5103 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5104 }
5105 SLAB_ATTR_RO(hwcache_align);
5106 
5107 #ifdef CONFIG_ZONE_DMA
5108 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5109 {
5110 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5111 }
5112 SLAB_ATTR_RO(cache_dma);
5113 #endif
5114 
5115 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5116 {
5117 	return sprintf(buf, "%u\n", s->usersize);
5118 }
5119 SLAB_ATTR_RO(usersize);
5120 
5121 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5122 {
5123 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5124 }
5125 SLAB_ATTR_RO(destroy_by_rcu);
5126 
5127 #ifdef CONFIG_SLUB_DEBUG
5128 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5129 {
5130 	return show_slab_objects(s, buf, SO_ALL);
5131 }
5132 SLAB_ATTR_RO(slabs);
5133 
5134 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5135 {
5136 	return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5137 }
5138 SLAB_ATTR_RO(total_objects);
5139 
5140 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5141 {
5142 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5143 }
5144 
5145 static ssize_t sanity_checks_store(struct kmem_cache *s,
5146 				const char *buf, size_t length)
5147 {
5148 	s->flags &= ~SLAB_CONSISTENCY_CHECKS;
5149 	if (buf[0] == '1') {
5150 		s->flags &= ~__CMPXCHG_DOUBLE;
5151 		s->flags |= SLAB_CONSISTENCY_CHECKS;
5152 	}
5153 	return length;
5154 }
5155 SLAB_ATTR(sanity_checks);
5156 
5157 static ssize_t trace_show(struct kmem_cache *s, char *buf)
5158 {
5159 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5160 }
5161 
5162 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5163 							size_t length)
5164 {
5165 	/*
5166 	 * Tracing a merged cache is going to give confusing results
5167 	 * as well as cause other issues like converting a mergeable
5168 	 * cache into an umergeable one.
5169 	 */
5170 	if (s->refcount > 1)
5171 		return -EINVAL;
5172 
5173 	s->flags &= ~SLAB_TRACE;
5174 	if (buf[0] == '1') {
5175 		s->flags &= ~__CMPXCHG_DOUBLE;
5176 		s->flags |= SLAB_TRACE;
5177 	}
5178 	return length;
5179 }
5180 SLAB_ATTR(trace);
5181 
5182 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5183 {
5184 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5185 }
5186 
5187 static ssize_t red_zone_store(struct kmem_cache *s,
5188 				const char *buf, size_t length)
5189 {
5190 	if (any_slab_objects(s))
5191 		return -EBUSY;
5192 
5193 	s->flags &= ~SLAB_RED_ZONE;
5194 	if (buf[0] == '1') {
5195 		s->flags |= SLAB_RED_ZONE;
5196 	}
5197 	calculate_sizes(s, -1);
5198 	return length;
5199 }
5200 SLAB_ATTR(red_zone);
5201 
5202 static ssize_t poison_show(struct kmem_cache *s, char *buf)
5203 {
5204 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5205 }
5206 
5207 static ssize_t poison_store(struct kmem_cache *s,
5208 				const char *buf, size_t length)
5209 {
5210 	if (any_slab_objects(s))
5211 		return -EBUSY;
5212 
5213 	s->flags &= ~SLAB_POISON;
5214 	if (buf[0] == '1') {
5215 		s->flags |= SLAB_POISON;
5216 	}
5217 	calculate_sizes(s, -1);
5218 	return length;
5219 }
5220 SLAB_ATTR(poison);
5221 
5222 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5223 {
5224 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5225 }
5226 
5227 static ssize_t store_user_store(struct kmem_cache *s,
5228 				const char *buf, size_t length)
5229 {
5230 	if (any_slab_objects(s))
5231 		return -EBUSY;
5232 
5233 	s->flags &= ~SLAB_STORE_USER;
5234 	if (buf[0] == '1') {
5235 		s->flags &= ~__CMPXCHG_DOUBLE;
5236 		s->flags |= SLAB_STORE_USER;
5237 	}
5238 	calculate_sizes(s, -1);
5239 	return length;
5240 }
5241 SLAB_ATTR(store_user);
5242 
5243 static ssize_t validate_show(struct kmem_cache *s, char *buf)
5244 {
5245 	return 0;
5246 }
5247 
5248 static ssize_t validate_store(struct kmem_cache *s,
5249 			const char *buf, size_t length)
5250 {
5251 	int ret = -EINVAL;
5252 
5253 	if (buf[0] == '1') {
5254 		ret = validate_slab_cache(s);
5255 		if (ret >= 0)
5256 			ret = length;
5257 	}
5258 	return ret;
5259 }
5260 SLAB_ATTR(validate);
5261 
5262 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5263 {
5264 	if (!(s->flags & SLAB_STORE_USER))
5265 		return -ENOSYS;
5266 	return list_locations(s, buf, TRACK_ALLOC);
5267 }
5268 SLAB_ATTR_RO(alloc_calls);
5269 
5270 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5271 {
5272 	if (!(s->flags & SLAB_STORE_USER))
5273 		return -ENOSYS;
5274 	return list_locations(s, buf, TRACK_FREE);
5275 }
5276 SLAB_ATTR_RO(free_calls);
5277 #endif /* CONFIG_SLUB_DEBUG */
5278 
5279 #ifdef CONFIG_FAILSLAB
5280 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5281 {
5282 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5283 }
5284 
5285 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5286 							size_t length)
5287 {
5288 	if (s->refcount > 1)
5289 		return -EINVAL;
5290 
5291 	s->flags &= ~SLAB_FAILSLAB;
5292 	if (buf[0] == '1')
5293 		s->flags |= SLAB_FAILSLAB;
5294 	return length;
5295 }
5296 SLAB_ATTR(failslab);
5297 #endif
5298 
5299 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5300 {
5301 	return 0;
5302 }
5303 
5304 static ssize_t shrink_store(struct kmem_cache *s,
5305 			const char *buf, size_t length)
5306 {
5307 	if (buf[0] == '1')
5308 		kmem_cache_shrink(s);
5309 	else
5310 		return -EINVAL;
5311 	return length;
5312 }
5313 SLAB_ATTR(shrink);
5314 
5315 #ifdef CONFIG_NUMA
5316 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5317 {
5318 	return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5319 }
5320 
5321 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5322 				const char *buf, size_t length)
5323 {
5324 	unsigned int ratio;
5325 	int err;
5326 
5327 	err = kstrtouint(buf, 10, &ratio);
5328 	if (err)
5329 		return err;
5330 	if (ratio > 100)
5331 		return -ERANGE;
5332 
5333 	s->remote_node_defrag_ratio = ratio * 10;
5334 
5335 	return length;
5336 }
5337 SLAB_ATTR(remote_node_defrag_ratio);
5338 #endif
5339 
5340 #ifdef CONFIG_SLUB_STATS
5341 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5342 {
5343 	unsigned long sum  = 0;
5344 	int cpu;
5345 	int len;
5346 	int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5347 
5348 	if (!data)
5349 		return -ENOMEM;
5350 
5351 	for_each_online_cpu(cpu) {
5352 		unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5353 
5354 		data[cpu] = x;
5355 		sum += x;
5356 	}
5357 
5358 	len = sprintf(buf, "%lu", sum);
5359 
5360 #ifdef CONFIG_SMP
5361 	for_each_online_cpu(cpu) {
5362 		if (data[cpu] && len < PAGE_SIZE - 20)
5363 			len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
5364 	}
5365 #endif
5366 	kfree(data);
5367 	return len + sprintf(buf + len, "\n");
5368 }
5369 
5370 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5371 {
5372 	int cpu;
5373 
5374 	for_each_online_cpu(cpu)
5375 		per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5376 }
5377 
5378 #define STAT_ATTR(si, text) 					\
5379 static ssize_t text##_show(struct kmem_cache *s, char *buf)	\
5380 {								\
5381 	return show_stat(s, buf, si);				\
5382 }								\
5383 static ssize_t text##_store(struct kmem_cache *s,		\
5384 				const char *buf, size_t length)	\
5385 {								\
5386 	if (buf[0] != '0')					\
5387 		return -EINVAL;					\
5388 	clear_stat(s, si);					\
5389 	return length;						\
5390 }								\
5391 SLAB_ATTR(text);						\
5392 
5393 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5394 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5395 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5396 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5397 STAT_ATTR(FREE_FROZEN, free_frozen);
5398 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5399 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5400 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5401 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5402 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5403 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5404 STAT_ATTR(FREE_SLAB, free_slab);
5405 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5406 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5407 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5408 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5409 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5410 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5411 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5412 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5413 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5414 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5415 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5416 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5417 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5418 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5419 #endif
5420 
5421 static struct attribute *slab_attrs[] = {
5422 	&slab_size_attr.attr,
5423 	&object_size_attr.attr,
5424 	&objs_per_slab_attr.attr,
5425 	&order_attr.attr,
5426 	&min_partial_attr.attr,
5427 	&cpu_partial_attr.attr,
5428 	&objects_attr.attr,
5429 	&objects_partial_attr.attr,
5430 	&partial_attr.attr,
5431 	&cpu_slabs_attr.attr,
5432 	&ctor_attr.attr,
5433 	&aliases_attr.attr,
5434 	&align_attr.attr,
5435 	&hwcache_align_attr.attr,
5436 	&reclaim_account_attr.attr,
5437 	&destroy_by_rcu_attr.attr,
5438 	&shrink_attr.attr,
5439 	&slabs_cpu_partial_attr.attr,
5440 #ifdef CONFIG_SLUB_DEBUG
5441 	&total_objects_attr.attr,
5442 	&slabs_attr.attr,
5443 	&sanity_checks_attr.attr,
5444 	&trace_attr.attr,
5445 	&red_zone_attr.attr,
5446 	&poison_attr.attr,
5447 	&store_user_attr.attr,
5448 	&validate_attr.attr,
5449 	&alloc_calls_attr.attr,
5450 	&free_calls_attr.attr,
5451 #endif
5452 #ifdef CONFIG_ZONE_DMA
5453 	&cache_dma_attr.attr,
5454 #endif
5455 #ifdef CONFIG_NUMA
5456 	&remote_node_defrag_ratio_attr.attr,
5457 #endif
5458 #ifdef CONFIG_SLUB_STATS
5459 	&alloc_fastpath_attr.attr,
5460 	&alloc_slowpath_attr.attr,
5461 	&free_fastpath_attr.attr,
5462 	&free_slowpath_attr.attr,
5463 	&free_frozen_attr.attr,
5464 	&free_add_partial_attr.attr,
5465 	&free_remove_partial_attr.attr,
5466 	&alloc_from_partial_attr.attr,
5467 	&alloc_slab_attr.attr,
5468 	&alloc_refill_attr.attr,
5469 	&alloc_node_mismatch_attr.attr,
5470 	&free_slab_attr.attr,
5471 	&cpuslab_flush_attr.attr,
5472 	&deactivate_full_attr.attr,
5473 	&deactivate_empty_attr.attr,
5474 	&deactivate_to_head_attr.attr,
5475 	&deactivate_to_tail_attr.attr,
5476 	&deactivate_remote_frees_attr.attr,
5477 	&deactivate_bypass_attr.attr,
5478 	&order_fallback_attr.attr,
5479 	&cmpxchg_double_fail_attr.attr,
5480 	&cmpxchg_double_cpu_fail_attr.attr,
5481 	&cpu_partial_alloc_attr.attr,
5482 	&cpu_partial_free_attr.attr,
5483 	&cpu_partial_node_attr.attr,
5484 	&cpu_partial_drain_attr.attr,
5485 #endif
5486 #ifdef CONFIG_FAILSLAB
5487 	&failslab_attr.attr,
5488 #endif
5489 	&usersize_attr.attr,
5490 
5491 	NULL
5492 };
5493 
5494 static const struct attribute_group slab_attr_group = {
5495 	.attrs = slab_attrs,
5496 };
5497 
5498 static ssize_t slab_attr_show(struct kobject *kobj,
5499 				struct attribute *attr,
5500 				char *buf)
5501 {
5502 	struct slab_attribute *attribute;
5503 	struct kmem_cache *s;
5504 	int err;
5505 
5506 	attribute = to_slab_attr(attr);
5507 	s = to_slab(kobj);
5508 
5509 	if (!attribute->show)
5510 		return -EIO;
5511 
5512 	err = attribute->show(s, buf);
5513 
5514 	return err;
5515 }
5516 
5517 static ssize_t slab_attr_store(struct kobject *kobj,
5518 				struct attribute *attr,
5519 				const char *buf, size_t len)
5520 {
5521 	struct slab_attribute *attribute;
5522 	struct kmem_cache *s;
5523 	int err;
5524 
5525 	attribute = to_slab_attr(attr);
5526 	s = to_slab(kobj);
5527 
5528 	if (!attribute->store)
5529 		return -EIO;
5530 
5531 	err = attribute->store(s, buf, len);
5532 #ifdef CONFIG_MEMCG
5533 	if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
5534 		struct kmem_cache *c;
5535 
5536 		mutex_lock(&slab_mutex);
5537 		if (s->max_attr_size < len)
5538 			s->max_attr_size = len;
5539 
5540 		/*
5541 		 * This is a best effort propagation, so this function's return
5542 		 * value will be determined by the parent cache only. This is
5543 		 * basically because not all attributes will have a well
5544 		 * defined semantics for rollbacks - most of the actions will
5545 		 * have permanent effects.
5546 		 *
5547 		 * Returning the error value of any of the children that fail
5548 		 * is not 100 % defined, in the sense that users seeing the
5549 		 * error code won't be able to know anything about the state of
5550 		 * the cache.
5551 		 *
5552 		 * Only returning the error code for the parent cache at least
5553 		 * has well defined semantics. The cache being written to
5554 		 * directly either failed or succeeded, in which case we loop
5555 		 * through the descendants with best-effort propagation.
5556 		 */
5557 		for_each_memcg_cache(c, s)
5558 			attribute->store(c, buf, len);
5559 		mutex_unlock(&slab_mutex);
5560 	}
5561 #endif
5562 	return err;
5563 }
5564 
5565 static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5566 {
5567 #ifdef CONFIG_MEMCG
5568 	int i;
5569 	char *buffer = NULL;
5570 	struct kmem_cache *root_cache;
5571 
5572 	if (is_root_cache(s))
5573 		return;
5574 
5575 	root_cache = s->memcg_params.root_cache;
5576 
5577 	/*
5578 	 * This mean this cache had no attribute written. Therefore, no point
5579 	 * in copying default values around
5580 	 */
5581 	if (!root_cache->max_attr_size)
5582 		return;
5583 
5584 	for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5585 		char mbuf[64];
5586 		char *buf;
5587 		struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5588 		ssize_t len;
5589 
5590 		if (!attr || !attr->store || !attr->show)
5591 			continue;
5592 
5593 		/*
5594 		 * It is really bad that we have to allocate here, so we will
5595 		 * do it only as a fallback. If we actually allocate, though,
5596 		 * we can just use the allocated buffer until the end.
5597 		 *
5598 		 * Most of the slub attributes will tend to be very small in
5599 		 * size, but sysfs allows buffers up to a page, so they can
5600 		 * theoretically happen.
5601 		 */
5602 		if (buffer)
5603 			buf = buffer;
5604 		else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
5605 			buf = mbuf;
5606 		else {
5607 			buffer = (char *) get_zeroed_page(GFP_KERNEL);
5608 			if (WARN_ON(!buffer))
5609 				continue;
5610 			buf = buffer;
5611 		}
5612 
5613 		len = attr->show(root_cache, buf);
5614 		if (len > 0)
5615 			attr->store(s, buf, len);
5616 	}
5617 
5618 	if (buffer)
5619 		free_page((unsigned long)buffer);
5620 #endif
5621 }
5622 
5623 static void kmem_cache_release(struct kobject *k)
5624 {
5625 	slab_kmem_cache_release(to_slab(k));
5626 }
5627 
5628 static const struct sysfs_ops slab_sysfs_ops = {
5629 	.show = slab_attr_show,
5630 	.store = slab_attr_store,
5631 };
5632 
5633 static struct kobj_type slab_ktype = {
5634 	.sysfs_ops = &slab_sysfs_ops,
5635 	.release = kmem_cache_release,
5636 };
5637 
5638 static int uevent_filter(struct kset *kset, struct kobject *kobj)
5639 {
5640 	struct kobj_type *ktype = get_ktype(kobj);
5641 
5642 	if (ktype == &slab_ktype)
5643 		return 1;
5644 	return 0;
5645 }
5646 
5647 static const struct kset_uevent_ops slab_uevent_ops = {
5648 	.filter = uevent_filter,
5649 };
5650 
5651 static struct kset *slab_kset;
5652 
5653 static inline struct kset *cache_kset(struct kmem_cache *s)
5654 {
5655 #ifdef CONFIG_MEMCG
5656 	if (!is_root_cache(s))
5657 		return s->memcg_params.root_cache->memcg_kset;
5658 #endif
5659 	return slab_kset;
5660 }
5661 
5662 #define ID_STR_LENGTH 64
5663 
5664 /* Create a unique string id for a slab cache:
5665  *
5666  * Format	:[flags-]size
5667  */
5668 static char *create_unique_id(struct kmem_cache *s)
5669 {
5670 	char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5671 	char *p = name;
5672 
5673 	BUG_ON(!name);
5674 
5675 	*p++ = ':';
5676 	/*
5677 	 * First flags affecting slabcache operations. We will only
5678 	 * get here for aliasable slabs so we do not need to support
5679 	 * too many flags. The flags here must cover all flags that
5680 	 * are matched during merging to guarantee that the id is
5681 	 * unique.
5682 	 */
5683 	if (s->flags & SLAB_CACHE_DMA)
5684 		*p++ = 'd';
5685 	if (s->flags & SLAB_CACHE_DMA32)
5686 		*p++ = 'D';
5687 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
5688 		*p++ = 'a';
5689 	if (s->flags & SLAB_CONSISTENCY_CHECKS)
5690 		*p++ = 'F';
5691 	if (s->flags & SLAB_ACCOUNT)
5692 		*p++ = 'A';
5693 	if (p != name + 1)
5694 		*p++ = '-';
5695 	p += sprintf(p, "%07u", s->size);
5696 
5697 	BUG_ON(p > name + ID_STR_LENGTH - 1);
5698 	return name;
5699 }
5700 
5701 static void sysfs_slab_remove_workfn(struct work_struct *work)
5702 {
5703 	struct kmem_cache *s =
5704 		container_of(work, struct kmem_cache, kobj_remove_work);
5705 
5706 	if (!s->kobj.state_in_sysfs)
5707 		/*
5708 		 * For a memcg cache, this may be called during
5709 		 * deactivation and again on shutdown.  Remove only once.
5710 		 * A cache is never shut down before deactivation is
5711 		 * complete, so no need to worry about synchronization.
5712 		 */
5713 		goto out;
5714 
5715 #ifdef CONFIG_MEMCG
5716 	kset_unregister(s->memcg_kset);
5717 #endif
5718 	kobject_uevent(&s->kobj, KOBJ_REMOVE);
5719 out:
5720 	kobject_put(&s->kobj);
5721 }
5722 
5723 static int sysfs_slab_add(struct kmem_cache *s)
5724 {
5725 	int err;
5726 	const char *name;
5727 	struct kset *kset = cache_kset(s);
5728 	int unmergeable = slab_unmergeable(s);
5729 
5730 	INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5731 
5732 	if (!kset) {
5733 		kobject_init(&s->kobj, &slab_ktype);
5734 		return 0;
5735 	}
5736 
5737 	if (!unmergeable && disable_higher_order_debug &&
5738 			(slub_debug & DEBUG_METADATA_FLAGS))
5739 		unmergeable = 1;
5740 
5741 	if (unmergeable) {
5742 		/*
5743 		 * Slabcache can never be merged so we can use the name proper.
5744 		 * This is typically the case for debug situations. In that
5745 		 * case we can catch duplicate names easily.
5746 		 */
5747 		sysfs_remove_link(&slab_kset->kobj, s->name);
5748 		name = s->name;
5749 	} else {
5750 		/*
5751 		 * Create a unique name for the slab as a target
5752 		 * for the symlinks.
5753 		 */
5754 		name = create_unique_id(s);
5755 	}
5756 
5757 	s->kobj.kset = kset;
5758 	err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5759 	if (err)
5760 		goto out;
5761 
5762 	err = sysfs_create_group(&s->kobj, &slab_attr_group);
5763 	if (err)
5764 		goto out_del_kobj;
5765 
5766 #ifdef CONFIG_MEMCG
5767 	if (is_root_cache(s) && memcg_sysfs_enabled) {
5768 		s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5769 		if (!s->memcg_kset) {
5770 			err = -ENOMEM;
5771 			goto out_del_kobj;
5772 		}
5773 	}
5774 #endif
5775 
5776 	kobject_uevent(&s->kobj, KOBJ_ADD);
5777 	if (!unmergeable) {
5778 		/* Setup first alias */
5779 		sysfs_slab_alias(s, s->name);
5780 	}
5781 out:
5782 	if (!unmergeable)
5783 		kfree(name);
5784 	return err;
5785 out_del_kobj:
5786 	kobject_del(&s->kobj);
5787 	goto out;
5788 }
5789 
5790 static void sysfs_slab_remove(struct kmem_cache *s)
5791 {
5792 	if (slab_state < FULL)
5793 		/*
5794 		 * Sysfs has not been setup yet so no need to remove the
5795 		 * cache from sysfs.
5796 		 */
5797 		return;
5798 
5799 	kobject_get(&s->kobj);
5800 	schedule_work(&s->kobj_remove_work);
5801 }
5802 
5803 void sysfs_slab_unlink(struct kmem_cache *s)
5804 {
5805 	if (slab_state >= FULL)
5806 		kobject_del(&s->kobj);
5807 }
5808 
5809 void sysfs_slab_release(struct kmem_cache *s)
5810 {
5811 	if (slab_state >= FULL)
5812 		kobject_put(&s->kobj);
5813 }
5814 
5815 /*
5816  * Need to buffer aliases during bootup until sysfs becomes
5817  * available lest we lose that information.
5818  */
5819 struct saved_alias {
5820 	struct kmem_cache *s;
5821 	const char *name;
5822 	struct saved_alias *next;
5823 };
5824 
5825 static struct saved_alias *alias_list;
5826 
5827 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5828 {
5829 	struct saved_alias *al;
5830 
5831 	if (slab_state == FULL) {
5832 		/*
5833 		 * If we have a leftover link then remove it.
5834 		 */
5835 		sysfs_remove_link(&slab_kset->kobj, name);
5836 		return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
5837 	}
5838 
5839 	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5840 	if (!al)
5841 		return -ENOMEM;
5842 
5843 	al->s = s;
5844 	al->name = name;
5845 	al->next = alias_list;
5846 	alias_list = al;
5847 	return 0;
5848 }
5849 
5850 static int __init slab_sysfs_init(void)
5851 {
5852 	struct kmem_cache *s;
5853 	int err;
5854 
5855 	mutex_lock(&slab_mutex);
5856 
5857 	slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
5858 	if (!slab_kset) {
5859 		mutex_unlock(&slab_mutex);
5860 		pr_err("Cannot register slab subsystem.\n");
5861 		return -ENOSYS;
5862 	}
5863 
5864 	slab_state = FULL;
5865 
5866 	list_for_each_entry(s, &slab_caches, list) {
5867 		err = sysfs_slab_add(s);
5868 		if (err)
5869 			pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5870 			       s->name);
5871 	}
5872 
5873 	while (alias_list) {
5874 		struct saved_alias *al = alias_list;
5875 
5876 		alias_list = alias_list->next;
5877 		err = sysfs_slab_alias(al->s, al->name);
5878 		if (err)
5879 			pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5880 			       al->name);
5881 		kfree(al);
5882 	}
5883 
5884 	mutex_unlock(&slab_mutex);
5885 	resiliency_test();
5886 	return 0;
5887 }
5888 
5889 __initcall(slab_sysfs_init);
5890 #endif /* CONFIG_SYSFS */
5891 
5892 /*
5893  * The /proc/slabinfo ABI
5894  */
5895 #ifdef CONFIG_SLUB_DEBUG
5896 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
5897 {
5898 	unsigned long nr_slabs = 0;
5899 	unsigned long nr_objs = 0;
5900 	unsigned long nr_free = 0;
5901 	int node;
5902 	struct kmem_cache_node *n;
5903 
5904 	for_each_kmem_cache_node(s, node, n) {
5905 		nr_slabs += node_nr_slabs(n);
5906 		nr_objs += node_nr_objs(n);
5907 		nr_free += count_partial(n, count_free);
5908 	}
5909 
5910 	sinfo->active_objs = nr_objs - nr_free;
5911 	sinfo->num_objs = nr_objs;
5912 	sinfo->active_slabs = nr_slabs;
5913 	sinfo->num_slabs = nr_slabs;
5914 	sinfo->objects_per_slab = oo_objects(s->oo);
5915 	sinfo->cache_order = oo_order(s->oo);
5916 }
5917 
5918 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
5919 {
5920 }
5921 
5922 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5923 		       size_t count, loff_t *ppos)
5924 {
5925 	return -EIO;
5926 }
5927 #endif /* CONFIG_SLUB_DEBUG */
5928